Optimization of poultry slaughterhouse wastewater treatment via advanced electrocoagulation using response surface methodology and a machine learning–based random forest–genetic algorithm hybrid model

Şule Eryürük; Kağan Eryürük

doi:10.25259/AJC_1225_2025

View/Download PDF

Buy Reprints

PDF

Translate this page into:

Original Article

ARTICLE IN PRESS

doi:

10.25259/AJC_1225_2025

Optimization of poultry slaughterhouse wastewater treatment via advanced electrocoagulation using response surface methodology and a machine learning–based random forest–genetic algorithm hybrid model

Şule Eryürük^a,, Kağan Eryürük^b,*

aDepartment of Industrial Engineering, KTO Karatay University, Akabe Mah. Alaaddin Kap, Konya, Turkey

bDepartment of Civil Engineering, Necmettin Erbakan University, Köyceğiz Mah. Demeç Sok. No:44/1, Konya, Turkey

*Corresponding author: E-mail address: kaganeryuruk@gmail.com (K. Eryürük)

Received: 2025-10-12, Accepted: 2026-02-09, Epub ahead of print: 2026-04-27,

Licence

This is an open-access article distributed under the terms of the Creative Commons Attribution-Non Commercial-Share Alike 4.0 License, which allows others to remix, transform, and build upon the work non-commercially, as long as the author is credited and the new creations are licensed under the identical terms.

Abstract

Chemical oxygen demand (COD), a major indicator of organic pollution in poultry slaughterhouse wastewater, must be effectively reduced prior to discharge. In this study, electrocoagulation was applied for COD removal, and the operating conditions were systematically optimized. The effects of current density, influent pH, flow rate, supporting electrolyte concentration, and hydrogen peroxide addition were investigated using response surface methodology (RSM) and a machine learning–based random forest–genetic algorithm (RF–GA) hybrid approach. RSM was employed to develop a statistically significant regression model (R² = 0.995, p <0.0001) describing the relationship between operating variables and COD concentration, although its predictive capability was constrained by assumptions of the polynomial model. In contrast, the RF–GA model captured nonlinear interactions more effectively and identified operating conditions yielding COD concentrations of approximately 460 mg/L (R² = 0.983), corresponding to a COD removal efficiency of 94.77%. Model interpretability was enhanced using shapley additive explanations (SHAP), which confirmed that influent pH and hydrogen peroxide concentration were the dominant factors governing COD reduction. In summary, the consistency between RSM, RF–GA, and SHAP analyses demonstrates the robustness and physical plausibility of the proposed optimization framework.

Keywords

Electrocoagulation

Genetic algorithm

Poultry slaughterhouse wastewater

Random forest

Response surface methodology

Random forest

Show Related Articles from PubMed

1. Introduction

Water is a vital natural resource for sustaining life and supporting sustainable development [1]. However, rapid population growth and urbanization have led to increased wastewater generation, placing increasing pressure on treatment infrastructure and receiving water bodies [2]. Among industrial effluents, poultry slaughterhouse wastewater is particularly challenging due to its large volume and high pollution load originating from slaughtering, meat processing, and intensive cleaning operations [3,4]. These effluents are typically characterized by elevated concentrations of organic matter, nitrogen, and phosphorus, posing serious environmental risks if inadequately treated [5].

Conventional treatment technologies for poultry slaughterhouse wastewater include aerobic [6], anaerobic [7], and combined biological processes [8]. Although effective under controlled conditions, biological treatments often suffer from operational instability, long hydraulic retention times, and reduced efficiency under shock organic loads. As a result, electrochemical treatment technologies—such as electrocoagulation, electrooxidation, and electrodialysis—have attracted increasing attention as alternative or complementary solutions [9,10]. Among these, electrocoagulation has emerged as a promising option due to its operational simplicity, low chemical consumption, reduced sludge generation, and ability to operate under ambient conditions. During electrocoagulation, sacrificial electrodes dissolve electrochemically, generating coagulant species that destabilize pollutants and promote floc formation and separation [11].

Several studies have demonstrated the feasibility of electrocoagulation for poultry slaughterhouse wastewater treatment. For example, Bazrafshan et al. [12] reported substantial chemical oxygen demand (COD) reduction using a bipolar aluminum electrode system, while Bayar et al. [13,14] achieved high removal efficiencies under different reactor configurations. Despite these encouraging outcomes, most reported studies focus on removal efficiency under limited operating conditions and rely heavily on trial-and-error experimentation, thereby restricting broader applicability and process scalability.

More importantly, the complex and nonlinear interactions among key electrocoagulation operating parameters—including current density, influent pH, flow rate, supporting electrolyte concentration, and oxidant dosage—remain insufficiently investigated for poultry slaughterhouse wastewater [15]. Recent research has therefore emphasized integrating electrochemical treatment processes with advanced statistical and data-driven modeling approaches to improve predictive accuracy and process understanding [16]. Machine learning techniques, such as artificial neural networks and adaptive neuro-fuzzy inference systems, have shown strong potential for capturing nonlinear behavior in wastewater treatment systems [17-19]. However, most existing studies apply either statistical optimization or machine learning independently, without exploiting their complementary strengths within a unified framework.

Recent hybrid modeling studies have demonstrated that combining statistical experimental design with machine learning algorithms significantly enhances the robustness and accuracy of optimization in complex wastewater treatment processes [20,21]. Such integrated approaches outperform conventional regression-based models, particularly under highly nonlinear conditions. Although previous studies have applied either statistical optimization [16,20] or machine learning–based prediction [17,21], none have systematically integrated response surface methodology (RSM) [22] with a random forest–genetic algorithm (RF–GA) hybrid framework specifically for the electrocoagulation treatment of poultry slaughterhouse wastewater. This integration enables both interpretability and robust global optimization, which remain unaddressed in the existing literature. To address this gap, the present study proposes a comprehensive optimization strategy for electrocoagulation-based treatment of poultry slaughterhouse wastewater. Response RSM is employed to systematically evaluate parameter interactions using a customized experimental design that accommodates an irregular design space. In parallel, a hybrid RF–GA approach is introduced to capture strong nonlinearities and identify globally optimal operating conditions. By comparing statistical and machine-learning–based optimization strategies, this study aims to enhance COD removal performance and provide a robust framework for electrocoagulation process optimization.

2. Materials and Methods

2.1. Wastewater and experimental details

The poultry slaughterhouse wastewater (psw) used in this study was collected from a poultry processing facility located in sakarya, türkiye. The wastewater was collected prior to entering the slaughterhouse’s conventional treatment plant. The wastewater had initial COD and biochemical oxygen demand (bod₅) concentrations of 8800 and 4250 mg/L, respectively; an initial pH of 7.1; and an initial electrical conductivity of 17.210 mS/cm. A continuous-flow operation controlled by a peristaltic pump was carried out to treat poultry slaughterhouse wastewater by electrocoagulation. The cathode was a cylindrical electrocoagulation reactor made of iron, with an effective volume of 675 mL, a height of 750 mm, and a diameter of 35 cm. The anodes were three iron rods, each with an outer diameter of 11 mm and a height of 720 mm, placed vertically inside the electrocoagulation reactor. The schematic representation of the experimental setup is shown in Figure 1. The anode and the cathode were connected to the positive and negative terminals of a direct current (DC) power supply, and 800 mL of poultry slaughterhouse wastewater (PSW) was pumped into the up flow reactor to evaluate the effects of current density, Na₂SO₄ concentration, flow rate, influent pH, and H₂O₂ concentration. Table 1 lists these variables and the values used in the experiments.

Schematic diagram of the continuous-flow electrocoagulation reactor used for poultry slaughterhouse wastewater treatment, consisting of a cylindrical iron cathode and three vertically aligned iron rod anodes connected to a DC power supply.

Table 1. Experimental factors and corresponding level settings used in the design of experiments (DoE) for electrocoagulation treatment of poultry slaughterhouse wastewater.

Name	Unit	L1	L2	L3	L4
A: Current density	mA/cm²	30	40	50	-
B: Influent pH		3	5	7.1	9
C: Supporting electrolyte (Na₂SO₄)	M	0	0.05	0.1	-
D: Flow Rate	mL/min	15	18	27	50
E: H₂O₂	M	0	0.1	0.15	0.2

Before each experiment, the pH was adjusted to the desired value using analytical-grade 1 N H₂SO₄ or NaOH (Merck, Darmstadt, Germany) to examine the effect of pH. In the experiments, analytical-grade Na₂SO₄ (Merck, Darmstadt, Germany) was used as the supporting electrolyte to investigate its effect [9].

Samples were collected to measure the initial COD of poultry slaughterhouse wastewater. Then a Statron T-25 power supply was used to apply current to the circuit for 90 min. A constant current was provided for the experiments. During the experiments, wastewater pH and conductivity were monitored using a Hanna 301 pH meter and a Radiometer Pioneer 30 conductivity meter, and samples collected at regular intervals were centrifuged and analyzed for COD by the closed-reflux method, according to Standard Methods for the Examination of Water and Wastewater [23]. All analyses were performed in duplicate, and the reported COD values represent the mean of replicate measurements. Diluted H₂SO₄ was used to polish the anodes and the cathode, which were iron rods and an iron shell, respectively, and purified water was used to rinse them before each experiment.

2.2. Effect of current density, flow rate, pH, supporting electrolyte, and H₂O₂ on COD concentration

The electrocoagulation process is influenced by several operational factors.

According to Faraday’s law, the current density regulates the release of Fe²⁺ ions, which promotes the formation of the coagulant within the system as given in Eq. 1[9].

(1)

m = \frac{i t M}{n F}

where m is the mass of dissolved iron (g Fe/cm), i is the current density (A/cm²), t is the reaction time (s), M is the iron molar mass (55.85 g/mol), n=2 the number of electrons in the oxidation step, and F Faraday’s constant (96,486 C/mol) [9].

Flow rate—closely linked to hydraulic retention time—is another key factor in electrocoagulation [24]. Reactants must reach the electrode surface before electrochemical conversion, after which the generated product (the coagulant) detaches from the electrode [24]. Consequently, higher flow rates enhance mass transfer and thereby improve COD removal efficiency.

Electrocoagulation performance is strongly influenced by the initial pH, which dictates the specific iron-hydroxide species formed during treatment [25].

According to the iron Pourbaix diagram, soluble iron hydroxide species predominate at both very low and very high pH values. At elevated pH, the hydroxide complex Fe(OH)₄⁻ becomes dominant, hindering floc formation [24]. Conversely, under highly acidic conditions, the scarcity of hydroxide ions limits iron hydroxide generation [24]. Therefore, operating near-neutral yields optimal COD removal efficiency.

Water’s electrolytic conductivity significantly affects the performance of electrochemical wastewater treatment. Introducing a supporting electrolyte lowers the resistance between the electrodes, boosting current efficiency and enabling optimal removal performance in an electrocoagulation reactor. H₂O₂ also plays a critical role. It participates in Fenton reactions to produce hydroxyl radicals, which are strong, non-selective oxidants [25]. These radicals promote oxidation, resulting in higher COD removal during the electrocoagulation process.

2.3. Definition of design of experiments (DoE)

The primary goal of experimental planning and execution is to derive reliable and unbiased conclusions. In this context, DoE serves as a systematic approach to structuring experiments efficiently while ensuring scientific rigor. By following a predefined experimental layout and analyzing the results using specific statistical methods, researchers can extract substantial insights from a limited number of experimental runs. This approach enables the simultaneous investigation of multiple variables, thereby reducing overall experimental costs. Moreover, DoE facilitates the identification of interactions between variables. Typically, the process begins with a broad examination of numerous factors, which is then refined to concentrate on the most influential variables [26].

2.4. Overview of response surface methodology(RSM)

RSM, introduced by Box and Wilson in 1951, is a mathematical and statistical technique designed to develop empirical models and optimize responses. It explores the relationship between multiple independent variables (inputs) and one or more dependent variables (responses), with the primary objective of optimizing the latter [27]. RSM relies on fitting polynomial equations to experimental data obtained through structured experimental designs. Although second-order (quadratic) polynomials are commonly used due to their effectiveness in capturing curvature in complex systems, higher-order polynomials can also be employed when necessary. In this approach, the independent variables are denoted as xi (where i ≥ 1), and the response variable is denoted as Y, which is the outcome to be optimized. The functional relationship between x_i and Y is typically expressed through a polynomial equation evaluated at various levels of x_i as presented in Eq. 2 [27].

(2)

y = f (x_{1}, x_{2}, x_{3}, \dots \dots, x_{i}, \dots \dots, x_{k}) + ε

ε represents the error term, or random noise, observed in response Y and accounts for unexplained variability in the system. The term x_i denotes the ith independent variable, while k represents the total number of input variables included in the model. The predicted or expected value of the response variable, known as the response surface, is modeled as a function of these independent variables and is generally expressed as in Eq. 3 [27].

(3)

E (y) = f (x_{1}, x_{2}, x_{3}, \dots \dots, x_{i}, \dots \dots, x_{k})

In most cases involving response surface methodology, the exact functional relationship between the independent variables and the response is initially unknown. Therefore, the first step in RSM involves assessing the underlying conditions to establish a suitable approximation that accurately represents the interaction between the response variable Y and the input variables. If the relationship among the input variables is strictly linear, the system can be adequately modeled by a first-order polynomial model, as shown in Eq. 4 [27].

(4)

y = β_{0} + β_{1} x_{1} + β_{2} x_{2} + \dots \dots \dots + β_{i} x_{i} + \dots \dots \dots \dots \dots \dots . + β_{k} x_{k} + ε

When curvature is present in the system, the response cannot be accurately captured by a linear model alone. In such cases, a higher-degree polynomial, typically a second-order model, is employed. This second-order approximation incorporates linear, quadratic (nonlinear), and interaction terms involving the independent variables x_i. These additional terms allow the model to better represent the underlying complexity of the response surface and capture the effects resulting from the interactions between pairs of variables as given in Eq. 5 [27].

(5)

y = β_{0} + \sum_{(i = 1)}^{k} β_{i} x_{i} + \sum_{(i = 1)}^{k} β_{(i i)} x_{i}^{2} + \sum \sum_{(i < j)} β_{i} j x_{i} x_{j} + ε

Analysis of variance (ANOVA) is a statistical method that decomposes the total variability observed in a dataset into distinct components associated with different sources of variation. The ANOVA table assesses the adequacy and significance of the fitted approximation models. It includes several key statistical metrics:

Sum of squares (SS): This quantifies the total variability in the data and partitions it into components attributable to different sources, such as treatments and error.
Degrees of freedom (df) — the number of independent values that can vary in the analysis. It reflects the amount of information available for estimating model parameters.
Mean square (MS) — calculated by dividing the sum of squares by the corresponding degrees of freedom — indicates the variance associated with each source. The treatment mean square reflects the variability between group means and is used to determine the significance of the model terms.
F-ratio (F value): the ratio of the treatment mean square to the error mean square; the F value provides a basis for hypothesis testing in ANOVA. It is compared against a critical value from the F-distribution, determined by the significance level (commonly α = 0.05) and the degrees of freedom.
p-value : This represents the probability of observing the test results under the null hypothesis. A p-value less than 0.05 typically indicates statistical significance, suggesting that differences observed among sample means are unlikely to be due to random chance.

Collectively, these elements of the ANOVA table provide a comprehensive statistical framework for evaluating the effectiveness and reliability of empirical models in experimental design.

Another important metric in response surface methodology is the desirability function, initially introduced by Myers et al. [27]. This function is employed as an objective function in multi-response optimization problems. It transforms each response variable to a scale ranging from 0 (indicating completely undesirable outcomes) to 1 (indicating the most desirable outcome), denoted di. The overall desirability is typically computed as the geometric mean of the individual desirability values for all responses. This composite measure provides a scalar value that can be maximized to determine the optimal operating conditions in numerical optimization procedures as presented in Eq. 6 [27].

(6)

D = {(d_{1} * d_{2} * \dots \dots \dots * d_{n})}^{\frac{1}{n}} = {(\prod_{i = 1}^{n} d_{i})}^{\frac{1}{n}}

2.5. Random forest model

The process involved generating multiple bootstrap samples from the original dataset, training a regression tree on each sample, and averaging the predictions of all trees to produce the final output. RF provides key advantages, such as evaluating the importance of input variables [4]. To ensure reliable performance, 90% of the data were randomly chosen for training and the remaining 10% for testing, thereby maintaining a representative dataset for model evaluation [4]. A random forest (RF) model was implemented in Python to predict COD removal efficiencies during electrocoagulation treatment of PSW.

2.6. Genetic algorithm

The genetic algorithm (GA), an evolutionary optimization method, mimics natural selection to efficiently explore complex, high-dimensional solution spaces [4]. By iteratively evaluating population fitness, GA preserves individuals with superior traits and balances crossover and mutation to maintain genetic diversity. Proper parameter settings—such as population size, crossover rate, and mutation rate—are critical; insufficient population size can cause premature convergence, while balanced crossover and mutation rates promote both exploration and retention of advantageous genes. This enables GA to identify near-optimal solutions even in uncertain or nonlinear environments [4].

2.7. Random forest – Genetic algorithm (RF–GA) hybrid approach

To enhance predictive accuracy and identify optimal operating conditions for COD minimization, a hybrid modeling and optimization framework was developed by integrating RF regression with a GA. The workflow of the RF–GA hybrid method is as follows:

Experimental data obtained from 36 runs, including current density, influent pH, flow rate, Na₂SO₄ concentration, and H₂O₂ concentration as independent variables, and COD as the response variable, were used to train the predictive model.

A random forest regression model was trained on the experimental dataset. RF was chosen due to its robustness in handling nonlinear relationships and its ability to rank features by importance, providing insights into the relative influence of each operational parameter on COD removal.

The predictive accuracy of the RF model was evaluated using cross-validation strategies, including k-fold CV and leave-one-out CV (LOO-CV). Performance metrics such as the coefficient of determination (R²), root mean square error (RMSE) and mean absolute error (MAE) were calculated.

Once the RF model was validated, it was embedded within a GA framework as the fitness function. In this setup:

The objective function was to minimize the COD concentration.
Decision variables included current density, influent pH, flow rate, and the concentrations of Na₂SO₄ and H₂O₂.
Constraints (bounds) were defined based on experimental ranges (e.g., current density: 30–50 mA/cm²; pH: 3–9; flow rate: 15–50 mL/min).
An initial population of candidate solutions was generated randomly within the defined parameter bounds.
Fitness values were calculated by inputting candidate solutions into the RF model to obtain predicted COD values.
Selection, crossover, and mutation operators were iteratively applied across multiple generations to explore the search space.
The best-performing individuals were retained as elites to ensure convergence toward the global optimum.
GA optimization provided the set of operational conditions corresponding to the predicted minimum COD. The convergence of fitness values across generations was monitored to ensure solution stability. Additionally, feature importance from the RF model and GA optimization curves were analyzed to interpret the predictive and optimization aspects of the hybrid approach.

The RF–GA hybrid approach successfully combined the nonlinear predictive capabilities of machine learning with the global optimization power of evolutionary algorithms, providing a reliable tool for determining the optimal operating conditions for COD removal in electrocoagulation.

3. Results and Discussion

3.1. Effect of operating parameters on COD removal

In the absence of direct material characterization, electrocoagulation mechanisms were inferred from established literature and observed process responses. Previous studies have demonstrated that the dissolution of iron electrodes during electrocoagulation leads to the generation of Fe²⁺ and Fe³⁺ species, followed by the formation of amorphous iron hydroxides such as Fe(OH)₂ and Fe(OH)₃, which play a key role in pollutant removal through charge neutralization, adsorption, and sweep flocculation mechanisms [28,29]. These iron hydroxide flocs provide a large reactive surface area, facilitating the aggregation and removal of organic contaminants from wastewater.

The experimental trends observed in this study support these well-established mechanisms. In particular, the strong dependence of COD removal efficiency on solution pH can be attributed to the pH-sensitive speciation of iron hydroxides and their coagulation behavior, as reported in previous electrocoagulation studies [28,29]. Similarly, the increase in COD removal with rising current density reflects enhanced iron dissolution and coagulant generation, which promote more effective destabilization and aggregation of organic matter [29]. Therefore, although no direct characterization analyses were conducted, the process-level responses obtained in this study provide indirect yet consistent evidence for iron-based electrocoagulation mechanisms governing COD removal performance.

3.2. Proposed models as functions for RSM

In this study, 36 experimental runs were conducted in a randomized order, following a custom design that enables a flexible experimental framework incorporating tailored models, categorical variables, and constrained or irregular design spaces. The analyses and optimization were carried out using Design-Expert software version 11.0.5.0 [30]. In this study, the independent variables—also referred to as operational parameters—included current density, pH, flow rate, hydrogen peroxide concentration (H₂O₂), and supporting electrolyte (Na₂SO₄). The COD value (Y) was selected as the dependent variable or response. The effectiveness of the experimental design was assessed by analyzing the extent to which COD reduction was achieved. The experimental findings are presented in Table 2.

Table 2. Experimental design matrix and corresponding COD concentrations obtained from electrocoagulation experiments conducted under varying operating conditions. Reported COD values represent the mean of duplicate measurements.

Run	Factor 1	Factor 2	Factor 3	Factor 4	Factor 5	Response 1
	A: Current density	B: Influent pH	C: Supporting electrolyte	D: Flow rate	E: H₂O₂	COD
	(mA/cm²)		(M)	(mL/min)	(M)	(mg/L)
1	30	7.1	0	18	0	1200
2	40	7.1	0	18	0	1075
3	50	7.1	0	18	0	1000
4	30	7.1	0.05	18	0	1524.16
5	40	7.1	0.05	18	0	1399.2
6	50	7.1	0.05	18	0	1323.52
7	30	7.1	0.1	18	0	2050.4
8	40	7.1	0.1	18	0	1600
9	50	7.1	0.1	18	0	1525.92
10	30	3	0	18	0	4964.96
11	40	3	0	18	0	4840
12	50	3	0	18	0	4764.32
13	30	5	0	18	0	4243.36
14	40	5	0	18	0	4118.4
15	50	5	0	18	0	4042.72
16	30	9	0	18	0	3495.36
17	40	9	0	18	0	3370.4
18	50	9	0	18	0	3294.72
19	30	7.1	0	15	0	1321.76
20	40	7.1	0	15	0	1196.8
21	50	7.1	0	15	0	1121.12
22	30	7.1	0	27	0	1128.16
23	40	7.1	0	27	0	1003.2
24	50	7.1	0	27	0	927.52
25	30	7.1	0	50	0	1056.88
26	40	7.1	0	50	0	931.92
27	50	7.1	0	50	0	856.24
28	30	3	0	18	0.1	915.2
29	40	3	0	18	0.1	827.2
30	50	3	0	18	0.1	626.56
31	30	3	0	18	0.15	783.2
32	40	3	0	18	0.15	695.2
33	50	3	0	18	0.15	494.56
34	30	3	0	18	0.2	616
35	40	3	0	18	0.2	528
36	50	3	0	18	0.2	327.36

Table 3 presents the ANOVA summary for this study at the 95% confidence level. The F-value for the model (6776.16) indicates statistical significance. There is only a 0.01% chance that an F-value this large could occur due to noise. P-values less than 0.0500 indicate that model terms are significant. In this case, B, C, D, E, AC, B², C², D², E², A²C, A²E, AC², B³, D³, and E³ are significant model terms. Values greater than 0.1000 indicate that the model terms are not statistically significant.

Table 3. ANOVA results for the reduced cubic RSM model using the Type III partial sum of squares method, indicating the statistical significance of model terms at a 95% confidence level.

Source	Sum of	df	Mean	F-value	p-value
Model	71504625	25	2860185	6776.1579	3.735E-18
A-Current density	94.9513	1	94.951299	0.2249522	0.6454789
B-Influent pH	2916113	1	2916112.9	6908.6585	1.553E-15
C-Supporting electrolyte	8491.151	1	8491.1507	20.116663	0.0011696
D-Flow Rate	7822.651	1	7822.6508	18.532898	0.0015492
E-H₂O₂	412097.9	1	412097.92	976.31469	2.648E-11
AB	15.30726	1	15.307265	0.0362649	0.8527798
AC	8014.728	1	8014.7277	18.987954	0.0014267
AD	24.59621	1	24.596208	0.0582717	0.8141265
AE	2089.237	1	2089.2369	4.9496796	0.0502849
A2	8.393663	1	8.3936633	0.0198857	0.890657
B2	6854277	1	6854277.4	16238.693	2.173E-17
C2	9053.193	1	9053.1927	21.448216	0.0009344
D2	8098.153	1	8098.1533	19.1856	0.0013771
E2	645288.1	1	645288.12	1528.7732	2.86E-12
A2B	20.74944	1	20.749437	0.0491582	0.8289968
A2C	11997.88	1	11997.884	28.424579	0.0003323
A2D	34.58242	1	34.582424	0.0819304	0.7805405
A2E	3311.763	1	3311.7633	7.8460068	0.01876
AB2	0.162816	1	0.1628158	0.0003857	0.9847169
AC2	9910.095	1	9910.0948	23.47833	0.000676
AD2	0.000461	1	0.0004613	1.093E-06	0.9991864
AE2	635.9808	1	635.98078	1.5067229	0.2477509
B3	8812509	1	8812509.1	20878.004	6.19E-18
D3	7502.763	1	7502.7629	17.775041	0.0017828
E3	416601.3	1	416601.33	986.98386	2.509E-11
Residual	4220.954	10	422.09538
Cor. Total	71508846	35

The corresponding mathematical model is expressed as follows:

y =-12066.69846 - 14.27023 * current density + 12904.08158 * influent pH + 21518.65266 * supporting electrolyte - 211.00035 * flow rate - 89700.51179 * H₂O₂ - 0.866471 * current density * influent pH - 952.88720 * current density * supporting electrolyte - 0.193916 * current density * flow rate + 282.47874 * current density * H₂O₂ + 0.051319 * current density² -2605.11295 * influent pH² + 1.26609E+05 * supporting electrolyte² + 7.09165 * flow rate² + 5.71215E+05 * H₂O₂² + 0.011395*current density² * influent pH + 13.93050 * current density² * supporting electrolyte + 0.002413 * current density² * flow rate - 4.29604 * current density² * H₂O₂ - 0.003213 * current density * influent pH² - 3235.95264 * current density * supporting electrolyte² + 6.83305E-06 * current density * flow rate² + 200.01354 * current density * H₂O₂² + 154.95342 * influent pH³ - 0.073038 * flow rate³ - 1.30279E+06 * H₂O₂³

Fit statistics for the evaluated model included standard deviation, mean, coefficient of variation (C.V.%), R², adjusted R², predicted R², and adequate precision. The standard deviation (20.54496) and mean (1810.814) indicate that the model predictions are highly consistent with the experimental data, as reflected by the low C.V. (1.13457%). The R² (0.999941) and adjusted R² (0.999793) values confirm the model’s strong ability to represent COD removal efficiency. Additionally, the predicted R² (0.995613) shows excellent agreement with the adjusted R², with a difference of less than 0.2, thereby verifying the model’s predictive reliability. The adequate precision value of 265.6637, which is greater than the threshold value of 4, demonstrates a high signal-to-noise ratio. Therefore, the model is considered statistically robust and appropriate for describing variations within the design space.

3.3. Assessment of model accuracy for RSM

Figure 2 provides a visual evaluation of the model’s predictive accuracy for COD values. The data points are closely aligned along the diagonal line, which represents a perfect prediction. This strong alignment indicates a high degree of correlation between the actual and predicted values, demonstrating that the regression model has good predictive performance. Additionally, the color gradient representing the range of COD values—from low (blue) to high (red)—shows a consistent spread along the fitted line, further supporting the adequacy and robustness of the model. This plot suggests that the developed model is both accurate and reliable in predicting COD removal outcomes under varying operational conditions.

Comparison between experimentally measured and RSM-predicted COD values for electrocoagulation of poultry slaughterhouse wastewater. Experimental COD values represent the mean of duplicate measurements, demonstrating good agreement between predicted and observed results within the experimental design space.

Figure 3(a) illustrates the interaction effect of current density (A) and influent pH (B) on COD removal performance. The plot reveals a pronounced curvature, indicating a significant interaction between these two variables. At lower pH and higher current densities, COD levels are substantially reduced, approaching the lower limit of the color scale (∼327 mg/L), suggesting enhanced degradation efficiency under these conditions. Conversely, higher pH levels and lower current densities correlate with elevated COD values, implying diminished treatment performance. It is worth noting that the predicted COD values become negative in certain regions of the plot, a physically implausible result that may indicate limitations in the model fit or extrapolation beyond the experimental bounds. Figure 3(a) underscores the importance of optimizing both pH and current density to achieve maximum COD removal in the given electrochemical treatment process.

Response surface plot showing the combined effect of current density and influent pH on COD concentration during electrocoagulation of poultry slaughterhouse wastewater (flow rate = 50 mL/min, Na2SO4= 0.05 M, H2O2 = 0 M). (b) Response surface plot illustrating the interactive effect of current density and supporting electrolyte (Na2SO4) concentration on COD concentration (influent pH = 4.74, flow rate = 50 mL/min, H2O2 = 0 M). (c) Response surface plot showing the combined influence of current density and flow rate on COD concentration during electrocoagulation (influent pH = 5.34, Na2SO4= 0.05 M, H2O2 = 0 M). (d) Response surface plot illustrating the interactive effect of current density and H2O2 concentration on COD concentration (influent pH = 3.18, flow rate = 41.25 mL/min, Na2SO4 = 0 M).

Figure 3(b) illustrates the combined effect of current density (A) and supporting electrolyte concentration (C) on COD levels during electrochemical treatment, with other variables held constant (influent pH = 4.74, flow rate = 50 mL/min, H₂O₂ concentration = 0). The surface remains relatively flat, exhibiting a gradual decrease in COD with increasing current density and supporting electrolyte concentration. This suggests that while both factors contribute to COD reduction, their influence is less pronounced under the given conditions. The limited curvature and narrow COD range in this plot imply a weak interaction between the two variables. Notably, the COD values are highest when both the current density and the supporting electrolyte concentration are low, and lowest when both factors are high. This trend is consistent with the expected behavior: higher current density increases the generation of oxidizing species, and the presence of a supporting electrolyte improves solution conductivity, thereby increasing treatment efficiency. However, the persistence of high COD values across the plot and the presence of slightly negative predictions again signal potential overfitting or extrapolation issues in the model, which should be interpreted cautiously. The results indicate a modest but synergistic benefit of increasing both the current density and the supporting electrolyte concentration to improve COD removal.

Figure 3(c) shows the interaction of current density (A) and flow rate (D) on COD levels during electrochemical treatment, with the other parameters held constant (influent pH = 5.34; supporting electrolyte = 0.05 M; H₂O₂ = 0). The response surface remains relatively flat, indicating only modest variation in COD across the tested ranges of current density and flow rate. A general downward trend in COD is observed as current density increases, reflecting the well-known beneficial effect of higher current densities on pollutant degradation through the generation of reactive species. However, the effect of flow rate is complex and nonlinear, with localized regions exhibiting slight improvements in COD removal at intermediate rates. The predicted COD values remain high across the surface, and some negative values reappear on the z-axis, likely due to extrapolation errors in the model. This raises concerns about the statistical robustness and model validity in these parameter ranges. Additionally, the minimal curvature and limited change in response suggest that, under the given conditions, neither current density nor flow rate significantly enhances COD removal over the studied range. The results indicate that although current density has a positive, albeit limited, impact, the flow rate does not exert a strong or consistent influence on COD reduction in this system configuration. Further optimization or extension of parameter ranges may be necessary to achieve more effective treatment performance.

Figure 3(d) illustrates the effect of two variables—current density (A; mA/cm²) and H₂O₂ concentration (E; M)—on (COD; mg/L) under fixed experimental conditions (influent pH = 3.18, no supporting electrolyte, and flow rate = 41.25 mL/min). The plot demonstrates a nonlinear interaction between current density and hydrogen peroxide concentration in their effects on COD levels. At lower H₂O₂ concentrations, increasing current density produces a marked decrease in COD, suggesting intensified electrochemical reactions and enhanced degradation of organic matter. At higher H₂O₂ concentrations, COD removal becomes less efficient, and COD values increase as current density decreases. This may be due to unreacted peroxide or to side reactions that inhibit COD removal efficiency. COD values range from 327.36 to 4964.96 mg/L, with the color gradient on the surface indicating regions of better and worse treatment performance (blue = lower COD, red = higher COD). This surface plot underscores the critical role of optimizing both current density and oxidant dose (H₂O₂) to achieve efficient COD reduction.

3.4. Model optimization and experimental validation for RSM

To ensure the validity of the developed regression model, it is essential to compare its predicted outcomes with the experimental data. Previous studies, such as those by Zularisam et al. [31] and Davarnejad et al. [32], have outlined procedures for validating regression models and assessing their predictive accuracy. In alignment with these approaches, the present study adopts a similar methodology for model verification.

As indicated by the software, the observed error is within an acceptable range. This is supported by the predicted R² (0.995613), which is in reasonable agreement with the adjusted R² (0.99979); the absolute difference is 0.00418 (0.42%), indicating a desirable model fit. Additionally, the model’s adequate precision is 265.6637, exceeding the threshold of 4. This suggests a strong signal-to-noise ratio, confirming that the developed model is statistically reliable and suitable for predictive purposes.

Through numerical optimization, optimal operating parameters were determined to minimize COD. During this stage, the objective was to keep all independent variables within their specified ranges, while minimizing the COD value.

In the present study, the condition that considered the variables in the model and possibly achieved the minimum COD value was selected as the optimal condition. Under these conditions, the COD value reaches a minimum of 439.175 mg/L when the following parameters are applied: current density 44.328 mA/cm², pH 4.967, flow rate 39.861 mL/min, H₂O₂ concentration 0.086 M, and Na₂SO₄ concentration 0.021 M.

3.5. Assessment of model accuracy for RF-GA

The RF model exhibited strong predictive capability with an average error of only 3.35%. Most predictions fell within ±5% error, confirming the model’s robustness. However, larger deviations (up to approximately 21.6%) were observed at very low COD levels, particularly under strongly acidic conditions (pH = 3) (Table 4). This can be attributed to data sparsity in that region of the design space, highlighting the need for additional experimental runs at low COD values to further refine the model’s accuracy. Overall, the model reliably captured COD removal trends across the investigated parameter space.

Table 4. Comparison of experimentally measured and RF–predicted COD values for electrocoagulation of poultry slaughterhouse wastewater under different operating conditions. Experimental COD values correspond to the mean of duplicate runs.

Run	Current density	Influent pH	Supporting electrolyte	Flow rate	H₂O₂	COD	Predicted COD
1	30	7.1	0	18	0	1200	1182.0857
2	40	7.1	0	18	0	1075	1092.4338
3	50	7.1	0	18	0	1000	1026.7779
4	30	7.1	0.05	18	0	1524.16	1489.4348
5	40	7.1	0.05	18	0	1399.2	1407.44
6	50	7.1	0.05	18	0	1323.52	1347.4381
7	30	7.1	0.1	18	0	2050.4	1868.3772
8	40	7.1	0.1	18	0	1600	1657.5919
9	50	7.1	0.1	18	0	1525.92	1556.7984
10	30	3	0	18	0	4964.96	4796.682
11	40	3	0	18	0	4840	4719.1804
12	50	3	0	18	0	4764.32	4619.786
13	30	5	0	18	0	4243.36	4243.4436
14	40	5	0	18	0	4118.4	4169.8272
15	50	5	0	18	0	4042.72	4075.2024
16	30	9	0	18	0	3495.36	3300.3143
17	40	9	0	18	0	3370.4	3248.2257
18	50	9	0	18	0	3294.72	3177.2465
19	30	7.1	0	15	0	1321.76	1275.3722
20	40	7.1	0	15	0	1196.8	1185.4411
21	50	7.1	0	15	0	1121.12	1118.333
22	30	7.1	0	27	0	1128.16	1106.1461
23	40	7.1	0	27	0	1003.2	1010.512
24	50	7.1	0	27	0	927.52	947.2627
25	30	7.1	0	50	0	1056.88	1053.1419
26	40	7.1	0	50	0	931.92	954.518
27	50	7.1	0	50	0	856.24	884.7766
28	30	3	0	18	0.1	915.2	875.4064
29	40	3	0	18	0.1	827.2	795.8412
30	50	3	0	18	0.1	626.56	645.1324
31	30	3	0	18	0.15	783.2	745.5536
32	40	3	0	18	0.15	695.2	670.23
33	50	3	0	18	0.15	494.56	532.1316
34	30	3	0	18	0.2	616	625.086
35	40	3	0	18	0.2	528	560.296
36	50	3	0	18	0.2	327.36	398.2132

Feature importance of the RF model (left), which highlights the relative contribution of each operational parameter to COD removal (Figure 4). Figure 4 shows that influent pH exerts the greatest influence on COD reduction, followed by hydrogen peroxide (H₂O₂) concentration; current density, electrolyte concentration (Na₂SO₄), and flow rate contribute less to the overall variance. This indicates that pH control and proper dosing of H₂O₂ are critical levers for optimizing electrocoagulation performance. The middle panel presents the relationship between actual and RF-predicted COD values. The clustering of points along the 1:1 reference line (red dashed) confirms the model’s high predictive accuracy and minimal systematic bias of the model, suggesting that the RF algorithm can capture the nonlinear interactions among the factors (Figure 4). Finally, Figure 4 (right panel) depicts the results of leave-one-out cross-validation (LOO-CV), yielding an R² of 0.983 and an RMSE of 183.16 mg/L. In this approach, each data point is iteratively excluded from the training set and used as a single test sample, ensuring that every observation is independently validated. These results validate not only the robustness of the RF model but also its strong generalization capability in small experimental datasets, thereby reinforcing its reliability as a surrogate model for subsequent GA-based optimization.

Performance evaluation of the RF model: (a) relative feature importance of operational parameters influencing COD reduction, (b) comparison between experimentally measured and RF-predicted COD values, and (c) LOO-CV results showing the predictive accuracy of the RF model.

To further evaluate the generalization capability of the RF model, a 90/10 train–test split was employed. The RF model achieved high training accuracy, with an R² of 0.996 and an RMSE of 1.664 (percent removal units). Importantly, comparable performance was observed on the test dataset, with an R² of 0.992 and an RMSE of 2.685% (removal units). The close agreement between training and test errors indicates that the model does not exhibit overfitting and is able to generalize effectively to unseen data. Moreover, consistency between these results and the leave-one-out cross-validation performance (R² = 0.983) further confirms the robustness and predictive reliability of the RF model, particularly given limited experimental datasets.

Complementary insights are provided by the SHAP (Shapley additive explanations) summary plot (Figure 5), which offers detailed interpretations of the RF model predictions at both the global and local levels. The SHAP analysis corroborates the feature importance results by identifying influent pH and influent H₂O₂ concentration as the dominant factors governing COD prediction. Specifically, lower pH levels and higher H₂O₂ doses are predominantly associated with negative SHAP values, indicating their strong contribution to COD reduction, whereas higher pH levels tend to increase predicted COD. Supporting electrolyte concentration exhibits a moderate effect, while current density and flow rate show relatively minor impacts on the model output. Overall, the consistency among RF feature importance, SHAP interpretation, and GA-identified optimal conditions confirms the physical plausibility of the model and reinforces confidence in the proposed hybrid RF–GA optimization framework.

SHAP summary plot showing the influence of input variables on COD prediction by the Random Forest model. Color represents feature magnitude, and SHAP values indicate the direction and strength of each variable’s contribution to the predicted COD.

GA optimization rapidly converged within the first 10 generations, stabilizing at a predicted COD of approximately 460 mg/L (Figure 6), which corresponds to a removal efficiency of 94.77%. Figure 6 indicates that the corresponding optimal conditions were determined to be a current density of 48.7 mA/cm², an influent pH of 6.77, an electrolyte concentration of 0.0186 M, a flow rate of 50 mL/min, and an H₂O₂ concentration of 0.185 M. These findings demonstrate that the RF–GA hybrid model effectively captures nonlinear interactions and identifies operating conditions that minimize COD levels.

Convergence behavior of the GA optimization showing the evolution of predicted minimum COD values over successive generations using the RF model as the fitness function.

The reduction in COD observed in this study is driven by the combined physicochemical mechanisms involved in electrocoagulation. Anodic dissolution of iron electrodes releases Fe²⁺ ions, which hydrolyze to form iron hydroxides (Fe(OH)₂ and Fe(OH)₃) that act as coagulants, destabilize pollutants, and promote their removal by sweep flocculation. Simultaneously, hydrogen peroxide enhances COD removal through Fenton-like reactions, with Fe²⁺ catalyzing H₂O₂ decomposition to produce reactive oxygen species, which oxidize complex organics into simpler, more readily removable forms. This synergy accounts for the dominant influence of influent pH and hydrogen peroxide concentration, as identified by both RSM and the RF–GA model, and demonstrates the consistency between statistical optimization and the underlying electrochemical mechanisms. Compared with other electrochemical treatments such as electrooxidation or electro-Fenton processes, the optimized electrocoagulation conditions offer effective COD abatement with simpler operation, lower chemical demand, and reduced process complexity, representing a practical and balanced approach for slaughterhouse wastewater treatment.

Figure 7 illustrates the performance of the GA across 10 independent runs, showing the evolution of the best COD (mg/L) values over successive generations. The primary objective of optimization is to minimize COD concentration.

Evolution of the best COD values obtained from 10 independent GA optimization runs as a function of generation number, illustrating convergence consistency and solution robustness.

In the early stages, considerable variability among runs is observed: some trajectories start above 700 mg/L, whereas others start below 500 mg/L. This variation reflects the stochastic nature of the initial populations and the diversity generated by their random initialization (Figure 7). Most runs exhibit a sharp decline in COD within the first 10–15 generations, often converging towards the 450–500 mg/L range. This demonstrates GA’s strong exploration capacity during the early search phase, enabling rapid improvements in solution quality. After approximately 15 generations, the curves begin to plateau, with the majority of runs stabilizing between 400–450 mg/L. The reduced frequency of major improvements in this interval suggests a transition from exploration to exploitation (Figure 7). By the 30th generation, nearly all runs have converged to approximately 400 mg/L, indicating that the GA consistently reaches near-global optimal solutions. Beyond this point, additional generations contribute little improvement, implying diminishing returns from extended iterations. These results confirm that the GA reliably converges to a similar low-COD region despite differences in initial conditions. The high variance in early generations highlights its exploration potential, while the stabilization phase underscores the algorithm’s robustness and solution maturity. Importantly, because no substantial improvements occur beyond the 30th generation, this point may serve as a practical stopping criterion for similar optimization problems (Figure 7).

A comparison of RSM-based optimization and the RF–GA hybrid framework reveals clear differences in predictive accuracy and optimization capability. The RSM model exhibited lower prediction errors than the RF–GA approach; this difference was confirmed as statistically significant by a paired Wilcoxon signed-rank test (p = 4.32 × 10⁻⁸). Bootstrap resampling further quantified the margin of error, showing a negative mean difference in absolute error (ΔError = Error_RSM − Error_RF = −43.44 mg/L) with a 95% confidence interval: −62.57 to −26.90 mg/L, indicating that RSM’s superior local predictive accuracy is robust and not due to random variation.

Despite this advantage, RSM-based optimization remains constrained by its quadratic approximation of the response surface, which may limit its ability to capture complex or irregular interactions among process variables. In contrast, the RF–GA framework demonstrated greater adaptability in nonlinear, multidimensional parameter spaces, identifying operating conditions that reduced COD concentrations to approximately 460 mg/L. This difference reflects the complementary nature of the two approaches: while RSM provides statistically efficient local modeling within structured experimental domains, RF–GA enables flexible global optimization through a data-driven surrogate model without assuming an explicit functional form [33-35].

4. Conclusions

This study demonstrated the applicability of electrocoagulation for treating poultry slaughterhouse wastewater within the operational ranges investigated. Experimental results confirmed that influent pH and hydrogen peroxide concentration exert the strongest influence on COD reduction, while current density, flow rate, and supporting electrolyte concentration play secondary but interacting roles.

RSM provided a statistically robust framework for identifying optimal operating conditions and interpreting factor interactions within the experimental design space. However, its performance was inherently constrained by polynomial assumptions, particularly under highly nonlinear conditions. In contrast, the RF–GA hybrid approach captured complex nonlinear relationships more effectively and identified alternative operating conditions associated with similarly low COD levels.

The comparative assessment indicates that statistical and machine-learning-based optimization methods offer complementary advantages: RSM ensures interpretability and experimental consistency, whereas RF–GA enhances adaptability and global optimization capabilities. Importantly, all conclusions are derived from experimentally validated data and remain limited to the operational ranges investigated in this study.

Overall, the combined use of RSM and data-driven optimization provides a balanced and reliable strategy for optimizing the electrocoagulation process, supporting its potential application to complex industrial wastewater streams with similar characteristics.

Future studies may extend the present work by expanding the experimental dataset to include a wider range of operating conditions and additional response variables, which would further enhance the generalization capability of machine learning–based models. The integration of external validation experiments at the optimal conditions identified by the RF–GA framework would provide additional confirmation of the model’s predictive reliability. Moreover, future research could explore multi-objective optimization by simultaneously considering COD removal efficiency, energy consumption, electrode consumption, and operational cost, thereby enabling a more comprehensive assessment of electrocoagulation performance. The incorporation of advanced deep learning architectures and hybrid physics-informed machine learning models may also improve the representation of complex electrochemical phenomena beyond the polynomial assumptions of RSM. Finally, scaling the proposed hybrid RSM–RF–GA framework to continuous-flow or pilot-scale systems would be a valuable step toward industrial implementation and real-world wastewater treatment applications.

CRediT authorship contribution statement

Şule Eryürük: Conceptualization, data curation, formal analysis, investigation, methodology, validation, visualization, writing – original draft, writing – review & editing; Kağan Eryürük: Conceptualization, data curation, formal analysis, investigation, methodology, validation, visualization, writing – original draft, writing – review & editing.

Declaration of competing interest

There are no conflicts of interest.

Data availability

Data related to this study can be made available upon request.

Declaration of generative AI and AI-assisted technologies in the writing process

The authors confirm that there was no use of artificial intelligence (AI)-assisted technology for assisting in the writing or editing of the manuscript and no images were manipulated using AI.

References

Fatima F., Du H., Kommalapati R.R.. Treatment of poultry slaughterhouse wastewater with membrane technologies: A review. Water. 2021;13:1905. https://doi.org/10.3390/w13141905
[Google Scholar]
Varjani S., Joshi R., Srivastava V.K., Ngo H.H., Guo W.. Treatment of wastewater from petroleum industry: Current practices and perspectives. Environmental Science and Pollution Research International. 2020;27:27172-27180. https://doi.org/10.1007/s11356-019-04725-x
[Google Scholar]
Dyosile P.A., Mdladla C., Njoya M., Basitere M., Ntwampe S.K.O., Kaskote E.. Assessment of an integrated and sustainable multistage system for the treatment of poultry slaughterhouse wastewater. Membranes. 2021;11:582. https://doi.org/10.3390/membranes11080582
[Google Scholar]
Kaskote E., Basitere M., Mshayisa V.V., Sheldon M.S.. Systematic review of poultry slaughterhouse wastewater treatment: Unveiling the potential of nanobubble technology. Water. 2024;16:1933. https://doi.org/10.3390/w16131933
[Google Scholar]
Meiramkulova K., Devrishov D., Zhumagulov M., Arystanova S., Karagoishin Z., Marzanova S., Kydyrbekova A., Mkilima T., Li J.. Performance of an integrated membrane process with electrochemical pre-treatment on poultry slaughterhouse wastewater purification. Membranes. 2020;10:256. https://doi.org/10.3390/membranes10100256
[Google Scholar]
Munir Baddour E., Farhoud N., Sharholy M., Mohammed I., Magid A.. Biological treatment of poultry slaughterhouses wastewater by using aerobic moving bed biofilm reactor. International Journal of Environmental Research and Public Health. 2016;3:96-106. https://doi.org/10.15739/irjpeh.16.013.
[Google Scholar]
Rajakumar R., Meenambal T., Saravanan P.M., Ananthanarayanan P.. Treatment of poultry slaughterhouse wastewater in hybrid upflow anaerobic sludge blanket reactor packed with pleated poly vinyl chloride rings. Bioresource Technology. 2012;103:116-122. https://doi.org/10.1016/j.biortech.2011.10.030
[Google Scholar]
Lopes C.L., de Assis T.M., Passig F.H., Lima Model A.N., Mees J.B.R., Cervantes F.J., Gotardo J.T., Gomes S.D.. Nitrogen removal from poultry slaughterhouse wastewater in anaerobic-anoxic-aerobic combined reactor: Integrated effect of recirculation rate and hydraulic retention time. Journal of Environmental Management. 2022;303:114162. https://doi.org/10.1016/j.jenvman.2021.114162
[Google Scholar]
Eryürük K., Eryürük Ş, Un U.T., Ogutveren U.B.. A design of experiment approach of cattle slaughterhouse wastewater treatment by electrocoagulation method. Desalination and Water Treatment. 2021;238:105-116. https://doi.org/10.5004/dwt.2021.27762
[Google Scholar]
Ngobeni P.V., Basitere M., Thole A.. Treatment of poultry slaughterhouse wastewater using electrocoagulation: A review. Water Practice and Technology. 2022;17:38-59. https://doi.org/10.2166/wpt.2021.108
[Google Scholar]
Eryuruk K., Tezcan Un U., Bakır Ogutveren U.. Electrochemical treatment of wastewaters from poultry slaughtering and processing by using iron electrodes. Journal of Cleaner Production. 2018;172:1089-1095. https://doi.org/10.1016/j.jclepro.2017.10.254
[Google Scholar]
Bazrafshan E., Kord Mostafapour F., Farzadkia M., Ownagh K.A., Mahvi A.H.. Slaughterhouse wastewater treatment by combined chemical coagulation and electrocoagulation process. PloS One. 2012;7:e40108. https://doi.org/10.1371/journal.pone.0040108
[Google Scholar]
Bayar S., Yıldız Y.Ş, Yılmaz A.E., İrdemez Ş. The effect of stirring speed and current density on removal efficiency of poultry slaughterhouse wastewater by electrocoagulation method. Desalination. 2011;280:103-107. https://doi.org/10.1016/j.desal.2011.06.061
[Google Scholar]
Bayar S., Yildiz Y.Ş, Yilmaz A.E., Koparal A.S.. The effect of initial pH on treatment of poultry slaughterhouse wastewater by electrocoagulation method. Desalination and Water Treatment. 2014;52:3047-3053. https://doi.org/10.1080/19443994.2013.800268
[Google Scholar]
Alver A., Altaş L.. Characterization and electrocoagulative treatment of landfill leachates: A statistical approach. Process Safety and Environmental Protection. 2017;111:102-111. https://doi.org/10.1016/j.psep.2017.04.021
[Google Scholar]
Baştürk E., Alver A.. Modeling azo dye removal by sono-fenton processes using response surface methodology and artificial neural network approaches. Journal of Environmental Management. 2019;248:109300. https://doi.org/10.1016/j.jenvman.2019.109300
[Google Scholar]
Alver A., Baştürk E., Tulun Ş., Şimşek İ.. Adaptive neuro‐fuzzy inference system modeling of 2,4‐dichlorophenol adsorption on wood‐based activated carbon. Environmental Progress & Sustainable Energy. 2020;39:e13413. https://doi.org/10.1002/ep.13413
[Google Scholar]
Nnaji P.C., Ume C.S., Obasi R.U., Anadebe V.C., Ezemagu I.G., Okeke B.U., Ude C.J., Onukwuli O.D.. Machine learning-based performance evaluation and sludge characterization studies of oxidized starch-aluminum electrode assisted by direct current treatment of dye laden wastewater. Results in Engineering. 2023;20:101576. https://doi.org/10.1016/j.rineng.2023.101576
[Google Scholar]
Alver A., Baştürk E., Kılıç A.. Development of adaptive neuro-fuzzy inference system model for predict trihalomethane formation potential in distribution network simulation test. Environmental Science and Pollution Research International. 2021;28:15870-15882. https://doi.org/10.1007/s11356-020-11801-0
[Google Scholar]
Alver A., Yılmaz B.A., Bilican B.K., Baştürk E., Kaya M., Işık M.. Carbamazepine adsorption onto giant macroporous silica and adaptive neuro-fuzzy inference system modeling. Arabian Journal for Science and Engineering. 2024;49:8509-8524. https://doi.org/10.1007/s13369-024-09032-3
[Google Scholar]
Argun M.E., Alver A., Karataş M.. Optimization of landfill leachate oxidation at extreme conditions and determination of micropollutants removal. Desalination and Water Treatment. 2017;90:130-138. https://doi.org/10.5004/dwt.2017.21241
[Google Scholar]
Okey-Onyesolu C.F., Chukwuma E.C., Okoye C.C., Onukwuli O.D.. Response surface methodology optimization of chito-protein synthesized from crab shell in treatment of abattoir wastewater. Heliyon. 2020;6:e05186. https://doi.org/10.1016/j.heliyon.2020.e05186
[Google Scholar]
American public health association, American water works association, Water environment federation. Standard methods for the examination of water and wastewater (22nd ed). Washington, DC.: American Public Health Association; 2012.
Tezcan Un U., Kandemir A., Erginel N., Ocal S.E.. Continuous electrocoagulation of cheese whey wastewater: An application of response surface methodology. Journal of Environmental Management. 2014;146:245-250. https://doi.org/10.1016/j.jenvman.2014.08.006
[Google Scholar]
Eryuruk K., Tezcan U.U., Bakır Ogutveren U.. Electrocoagulation in a plug flow reactor: The treatment of cattle abattoir wastewater by iron rod anodes. International Journal of Environmental Research. 2014;8:461-468. http://www.ijer.ir/article_738_36.html.
[Google Scholar]
Tague N.R.. The quality toolbox. Wisconsin.: ASQ Quality Press; 2005.
Myers R.H., Montgomery D.C., Anderson-Cook C.M.. Response surface methodology: Process and product optimization using designed experiments (fourth ed). New Jersey: John Wiley & Sons; 2016.
Mollah M.Y., Schennach R., Parga J.R., Cocke D.L.. Electrocoagulation (EC) — science and applications. Journal of Hazardous Materials. 2001;84:29-41. https://doi.org/10.1016/s0304-3894(01)00176-5
[Google Scholar]
Holt P.K., Barton G.W., Mitchell C.A.. The future for electrocoagulation as a localised water treatment technology. Chemosphere. 2005;59:355-367. https://doi.org/10.1016/j.chemosphere.2004.10.023
[Google Scholar]
Statease, 2016. Convert coded response surface model to actual. [accessed 2025 May 20].
Zularisam A.W., Ismail A.F., Salim M.R., Sakinah M., Matsuura T.. Application of coagulation–ultrafiltration hybrid process for drinking water treatment: Optimization of operating conditions using experimental design. Separation and Purification Technology. 2009;65:193-210. https://doi.org/10.1016/j.seppur.2008.10.018
[Google Scholar]
Davarnejad R., Sahraei A.. Industrial wastewater treatment using an electrochemical technique: An optimized process. Desalination and Water Treatment. 2016;57:9622-9634. https://doi.org/10.1080/19443994.2015.1030776
[Google Scholar]
Wang K., Mao Y., Wang C., Ke Q., Zhao M., Wang Q.. Application of a combined response surface methodology (RSM)-artificial neural network (ANN) for multiple target optimization and prediction in a magnetic coagulation process for secondary effluent from municipal wastewater treatment plants. Environmental Science and Pollution Research. 2022;29:36075-36087. https://doi.org/10.1007/s11356-021-18060-7
[Google Scholar]
Zain N.S., Shaaban I.A., Zafar F., Asif H.M., Tariq M., Khan M.A., Akhtar N., Kainat H., Assiri M.A.. Machine learning integration with response surface methodology to enhance the removal efficacy of arsenate (V) through sulfur-functionalized mxene coated QPPO/PVA AEM. Journal of Environmental Management. 2024;371:123211. https://doi.org/10.1016/j.jenvman.2024.123211
[Google Scholar]
Kang J.K., Lee Y.J., Son C.Y., Park S.J., Lee C.G.. Alternative assessment of machine learning to polynomial regression in response surface methodology for predicting decolorization efficiency in textile wastewater treatment. Chemosphere. 2025;370:143996. https://doi.org/10.1016/j.chemosphere.2024.143996
[Google Scholar]

1. Introduction

2. Materials and Methods

2.1. Wastewater and experimental details
2.2. Effect of current density, flow rate, pH, supporting electrolyte, and H2O2 on COD concentration
2.3. Definition of design of experiments (DoE)
2.4. Overview of response surface methodology(RSM)
2.5. Random forest model
2.6. Genetic algorithm
2.7. Random forest – Genetic algorithm (RF–GA) hybrid approach

3. Results and Discussion

3.1. Effect of operating parameters on COD removal
3.2. Proposed models as functions for RSM
3.3. Assessment of model accuracy for RSM
3.4. Model optimization and experimental validation for RSM
3.5. Assessment of model accuracy for RF-GA

4. Conclusions

Fulltext Views
373

PDF downloads
499

View/Download PDF
Download Citations

BibTeX
RIS

Show Sections

[1] Fatima F., Du H., Kommalapati R.R.. Treatment of poultry slaughterhouse wastewater with membrane technologies: A review. Water. 2021;13:1905. https://doi.org/10.3390/w13141905
[Google Scholar]

[2] Varjani S., Joshi R., Srivastava V.K., Ngo H.H., Guo W.. Treatment of wastewater from petroleum industry: Current practices and perspectives. Environmental Science and Pollution Research International. 2020;27:27172-27180. https://doi.org/10.1007/s11356-019-04725-x
[Google Scholar]

[3] Dyosile P.A., Mdladla C., Njoya M., Basitere M., Ntwampe S.K.O., Kaskote E.. Assessment of an integrated and sustainable multistage system for the treatment of poultry slaughterhouse wastewater. Membranes. 2021;11:582. https://doi.org/10.3390/membranes11080582
[Google Scholar]

[4] Kaskote E., Basitere M., Mshayisa V.V., Sheldon M.S.. Systematic review of poultry slaughterhouse wastewater treatment: Unveiling the potential of nanobubble technology. Water. 2024;16:1933. https://doi.org/10.3390/w16131933
[Google Scholar]

[5] Meiramkulova K., Devrishov D., Zhumagulov M., Arystanova S., Karagoishin Z., Marzanova S., Kydyrbekova A., Mkilima T., Li J.. Performance of an integrated membrane process with electrochemical pre-treatment on poultry slaughterhouse wastewater purification. Membranes. 2020;10:256. https://doi.org/10.3390/membranes10100256
[Google Scholar]

[6] Munir Baddour E., Farhoud N., Sharholy M., Mohammed I., Magid A.. Biological treatment of poultry slaughterhouses wastewater by using aerobic moving bed biofilm reactor. International Journal of Environmental Research and Public Health. 2016;3:96-106. https://doi.org/10.15739/irjpeh.16.013.
[Google Scholar]

[7] Rajakumar R., Meenambal T., Saravanan P.M., Ananthanarayanan P.. Treatment of poultry slaughterhouse wastewater in hybrid upflow anaerobic sludge blanket reactor packed with pleated poly vinyl chloride rings. Bioresource Technology. 2012;103:116-122. https://doi.org/10.1016/j.biortech.2011.10.030
[Google Scholar]

[8] Lopes C.L., de Assis T.M., Passig F.H., Lima Model A.N., Mees J.B.R., Cervantes F.J., Gotardo J.T., Gomes S.D.. Nitrogen removal from poultry slaughterhouse wastewater in anaerobic-anoxic-aerobic combined reactor: Integrated effect of recirculation rate and hydraulic retention time. Journal of Environmental Management. 2022;303:114162. https://doi.org/10.1016/j.jenvman.2021.114162
[Google Scholar]

[9] Eryürük K., Eryürük Ş, Un U.T., Ogutveren U.B.. A design of experiment approach of cattle slaughterhouse wastewater treatment by electrocoagulation method. Desalination and Water Treatment. 2021;238:105-116. https://doi.org/10.5004/dwt.2021.27762
[Google Scholar]

[10] Ngobeni P.V., Basitere M., Thole A.. Treatment of poultry slaughterhouse wastewater using electrocoagulation: A review. Water Practice and Technology. 2022;17:38-59. https://doi.org/10.2166/wpt.2021.108
[Google Scholar]

[11] Eryuruk K., Tezcan Un U., Bakır Ogutveren U.. Electrochemical treatment of wastewaters from poultry slaughtering and processing by using iron electrodes. Journal of Cleaner Production. 2018;172:1089-1095. https://doi.org/10.1016/j.jclepro.2017.10.254
[Google Scholar]

[12] Bazrafshan E., Kord Mostafapour F., Farzadkia M., Ownagh K.A., Mahvi A.H.. Slaughterhouse wastewater treatment by combined chemical coagulation and electrocoagulation process. PloS One. 2012;7:e40108. https://doi.org/10.1371/journal.pone.0040108
[Google Scholar]

[13] Bayar S., Yıldız Y.Ş, Yılmaz A.E., İrdemez Ş. The effect of stirring speed and current density on removal efficiency of poultry slaughterhouse wastewater by electrocoagulation method. Desalination. 2011;280:103-107. https://doi.org/10.1016/j.desal.2011.06.061
[Google Scholar]

[14] Bayar S., Yildiz Y.Ş, Yilmaz A.E., Koparal A.S.. The effect of initial pH on treatment of poultry slaughterhouse wastewater by electrocoagulation method. Desalination and Water Treatment. 2014;52:3047-3053. https://doi.org/10.1080/19443994.2013.800268
[Google Scholar]

[15] Alver A., Altaş L.. Characterization and electrocoagulative treatment of landfill leachates: A statistical approach. Process Safety and Environmental Protection. 2017;111:102-111. https://doi.org/10.1016/j.psep.2017.04.021
[Google Scholar]

[16] Baştürk E., Alver A.. Modeling azo dye removal by sono-fenton processes using response surface methodology and artificial neural network approaches. Journal of Environmental Management. 2019;248:109300. https://doi.org/10.1016/j.jenvman.2019.109300
[Google Scholar]

[17] Alver A., Baştürk E., Tulun Ş., Şimşek İ.. Adaptive neuro‐fuzzy inference system modeling of 2,4‐dichlorophenol adsorption on wood‐based activated carbon. Environmental Progress & Sustainable Energy. 2020;39:e13413. https://doi.org/10.1002/ep.13413
[Google Scholar]

[18] Nnaji P.C., Ume C.S., Obasi R.U., Anadebe V.C., Ezemagu I.G., Okeke B.U., Ude C.J., Onukwuli O.D.. Machine learning-based performance evaluation and sludge characterization studies of oxidized starch-aluminum electrode assisted by direct current treatment of dye laden wastewater. Results in Engineering. 2023;20:101576. https://doi.org/10.1016/j.rineng.2023.101576
[Google Scholar]

[19] Alver A., Baştürk E., Kılıç A.. Development of adaptive neuro-fuzzy inference system model for predict trihalomethane formation potential in distribution network simulation test. Environmental Science and Pollution Research International. 2021;28:15870-15882. https://doi.org/10.1007/s11356-020-11801-0
[Google Scholar]

[20] Alver A., Yılmaz B.A., Bilican B.K., Baştürk E., Kaya M., Işık M.. Carbamazepine adsorption onto giant macroporous silica and adaptive neuro-fuzzy inference system modeling. Arabian Journal for Science and Engineering. 2024;49:8509-8524. https://doi.org/10.1007/s13369-024-09032-3
[Google Scholar]

[21] Argun M.E., Alver A., Karataş M.. Optimization of landfill leachate oxidation at extreme conditions and determination of micropollutants removal. Desalination and Water Treatment. 2017;90:130-138. https://doi.org/10.5004/dwt.2017.21241
[Google Scholar]

[22] Okey-Onyesolu C.F., Chukwuma E.C., Okoye C.C., Onukwuli O.D.. Response surface methodology optimization of chito-protein synthesized from crab shell in treatment of abattoir wastewater. Heliyon. 2020;6:e05186. https://doi.org/10.1016/j.heliyon.2020.e05186
[Google Scholar]

[23] American public health association, American water works association, Water environment federation. Standard methods for the examination of water and wastewater (22nd ed). Washington, DC.: American Public Health Association; 2012.

[24] Tezcan Un U., Kandemir A., Erginel N., Ocal S.E.. Continuous electrocoagulation of cheese whey wastewater: An application of response surface methodology. Journal of Environmental Management. 2014;146:245-250. https://doi.org/10.1016/j.jenvman.2014.08.006
[Google Scholar]

[25] Eryuruk K., Tezcan U.U., Bakır Ogutveren U.. Electrocoagulation in a plug flow reactor: The treatment of cattle abattoir wastewater by iron rod anodes. International Journal of Environmental Research. 2014;8:461-468. http://www.ijer.ir/article_738_36.html.
[Google Scholar]

[26] Tague N.R.. The quality toolbox. Wisconsin.: ASQ Quality Press; 2005.

[27] Myers R.H., Montgomery D.C., Anderson-Cook C.M.. Response surface methodology: Process and product optimization using designed experiments (fourth ed). New Jersey: John Wiley & Sons; 2016.

[28] Mollah M.Y., Schennach R., Parga J.R., Cocke D.L.. Electrocoagulation (EC) — science and applications. Journal of Hazardous Materials. 2001;84:29-41. https://doi.org/10.1016/s0304-3894(01)00176-5
[Google Scholar]

[29] Holt P.K., Barton G.W., Mitchell C.A.. The future for electrocoagulation as a localised water treatment technology. Chemosphere. 2005;59:355-367. https://doi.org/10.1016/j.chemosphere.2004.10.023
[Google Scholar]

[30] Statease, 2016. Convert coded response surface model to actual. [accessed 2025 May 20].

[31] Zularisam A.W., Ismail A.F., Salim M.R., Sakinah M., Matsuura T.. Application of coagulation–ultrafiltration hybrid process for drinking water treatment: Optimization of operating conditions using experimental design. Separation and Purification Technology. 2009;65:193-210. https://doi.org/10.1016/j.seppur.2008.10.018
[Google Scholar]

[32] Davarnejad R., Sahraei A.. Industrial wastewater treatment using an electrochemical technique: An optimized process. Desalination and Water Treatment. 2016;57:9622-9634. https://doi.org/10.1080/19443994.2015.1030776
[Google Scholar]

[33] Wang K., Mao Y., Wang C., Ke Q., Zhao M., Wang Q.. Application of a combined response surface methodology (RSM)-artificial neural network (ANN) for multiple target optimization and prediction in a magnetic coagulation process for secondary effluent from municipal wastewater treatment plants. Environmental Science and Pollution Research. 2022;29:36075-36087. https://doi.org/10.1007/s11356-021-18060-7
[Google Scholar]

[34] Zain N.S., Shaaban I.A., Zafar F., Asif H.M., Tariq M., Khan M.A., Akhtar N., Kainat H., Assiri M.A.. Machine learning integration with response surface methodology to enhance the removal efficacy of arsenate (V) through sulfur-functionalized mxene coated QPPO/PVA AEM. Journal of Environmental Management. 2024;371:123211. https://doi.org/10.1016/j.jenvman.2024.123211
[Google Scholar]

[35] Kang J.K., Lee Y.J., Son C.Y., Park S.J., Lee C.G.. Alternative assessment of machine learning to polynomial regression in response surface methodology for predicting decolorization efficiency in textile wastewater treatment. Chemosphere. 2025;370:143996. https://doi.org/10.1016/j.chemosphere.2024.143996
[Google Scholar]

Optimization of poultry slaughterhouse wastewater treatment via advanced electrocoagulation using response surface methodology and a machine learning–based random forest–genetic algorithm hybrid model