Sustainable Wastewater Treatment and Water Reuse via Electrochemical Advanced Oxidation of Trypan Blue Using Boron-Doped Diamond Anode: XGBoost-Based Performance Prediction

Abstract

Azo dyes are widely used in the textile industry due to their vibrant colors and chemical stability; however, wastewater containing these dyes poses significant environmental and health risks due to their toxic, persistent, and potentially carcinogenic properties. In this study, the treatment of wastewater containing trypan blue dye was investigated using the electrooxidation process with boron-doped diamond electrodes, and the efficiency of the process was modeled through the Extreme Gradient Boosting (XGBoost) algorithm. In the experimental phase, the effects of key operational parameters, including current density, pH, electrolysis time, and supporting electrolyte concentration, on TB dye removal efficiency were systematically evaluated. Based on the experimental data obtained, a machine learning-based XGBoost prediction model was developed, and hyperparameter optimization was performed to enhance its predictive performance. The model achieved high accuracy (R² = 0.996 for training and 0.954 for testing) and yielded low error metrics (RMSE and MAE), confirming its reliability in predicting removal efficiency. This study presents an integrated and data-driven approach for improving the efficiency and sustainability of electrooxidation processes and offers an environmentally friendly and effective method for the treatment of azo dye-contaminated wastewater.

1. Introduction

Water is an indispensable resource for life, and the preservation of water resources has become more critical than ever due to factors such as the global climate crisis, increasing industrialization, and prolonged droughts. The textile industry is a prominent sector in this context, owing to its intensive water usage and substantial wastewater generation during production processes [1]. Among the most significant chemical compounds used in this industry are azo dyes, which are extensively applied in dyeing operations [2]. To enhance dye fixation on fabrics, the molecular structures of these dyes are modified, resulting in various forms such as acid dyes, direct dyes, reactive dyes, and disperse dyes [3]. Azo dyes are synthetic compounds characterized by one or more azo (-N=N-) groups and are widely utilized in the textile dyeing processes [4]. Depending on the number of azo groups in their structure, these compounds are categorized as monoazo, diazo, or triazo dyes [5]. In addition to being linked to benzene or naphthalene rings, they can also be bonded to heterocyclic aromatics and aliphatic groups [6].

Azo dye pollution originating from the textile industry poses severe threats to aquatic ecosystems. Due to their resistance to biodegradation, these dyes exhibit environmental persistence, leading to long-term ecological damage. Several studies have reported that certain azo dyes exhibit carcinogenic and mutagenic properties, posing serious risks to human health [5,7]. These compounds tend to accumulate in aquatic environments, where they exert toxic effects, contribute to water pollution, and diminish biodiversity [5,8]. Therefore, it is essential to develop innovative and efficient treatment technologies for the removal of azo dyes from industrial wastewater.

In addition to conventional physico-chemical methods such as biological treatment, chemical oxidation, coagulation-flocculation [9], membrane filtration [10], and ion exchange [11], electrochemical methods—including electro-Fenton [12], anodic oxidation, and electrocoagulation [13]—have gained significant attention. Electrochemical techniques have emerged as promising alternatives in recent years due to their high removal efficiencies and environmentally benign characteristics. In this regard, the electrooxidation process stands out as an advanced oxidation technique capable of mineralizing organic pollutants and enhancing water reusability.

Electrooxidation is an electrochemical treatment method that produces strong oxidizing species at the anode surface for the removal of organic pollutants [14]. The efficiency of the electrooxidation process can be optimized by adjusting variables such as electrode type, current density, electrolyte composition, and electrolysis time [15]. This process is generally carried out with corrosion-resistant anode materials such as graphite [16], coated titanium [17], platinum [18], and boron-doped diamond (BDD) [19]. Among these electrodes, BDD stands out due to its high oxidation potential, wide electrochemical stability range, and chemical inertness.

Accordingly, in aqueous media, BDD enables operation at anode potentials where strong oxidants such as hydroxyl radicals (•OH; E°(•OH/H₂O) ≈ 2.80 V vs. SHE) are formed, consistent with its non-active surface that minimizes passivation [20]. BDD further exhibits a wide aqueous potential window of approximately −1.25 to +2.3 V vs. SHE, providing access to the anodic regime required for non-selective oxidation [21]. The oxygen evolution reaction (OER) typically onsets at ~1.5–1.8 V vs. Ag/AgCl for thin-film BDD, favoring radical generation over parasitic oxygen evolution [22]. These characteristics translate into high mineralization efficiency and durability relative to other corrosion-resistant anodes in wastewater applications [23]. Although BDD is more expensive than other non-sacrificial anodes, this limitation can be mitigated at practical scale by operating in energy-efficient windows (e.g., optimized current density/hydrodynamics) and through reactor/energy integration (e.g., renewable coupling, hybrid polishing), as highlighted in recent critical assessments [24].

Highly reactive oxidants, such as •OH, are effectively generated on the BDD electrode surface, which convert complex organic compounds into biodegradable intermediates or fully mineralized compounds Equations (1) and (2) [25,26]. Furthermore, they do not show electrode passivation or surface degradation even at high current densities, offering significant advantages in terms of long-term stability and low maintenance requirements. Thanks to these superior properties, BDD electrodes are considered the most suitable anode material for the removal of complex organic dyes via advanced oxidation, providing high efficiency, durability, and energy efficiency [15,27,28]. The advantages of the electrooxidation process, such as its general high efficiency, lack of chemical requirements, and ease of system integration, make this method stand out among sustainable wastewater treatment technologies [22,29,30,31,32].

BDD + H₂O → BDD (∙OH) + H⁺ + e⁻

(1)

BDD(∙OH) + TB Dye → CO₂ + H₂O + H⁺ + e⁻

(2)

Studies employing BDD anodes have demonstrated that process performance is significantly influenced by parameters including electrolyte type, current density, and pH.

In recent years, the application of machine learning (ML) techniques for optimizing environmental treatment processes such as electrooxidation has become increasingly prevalent. ML models are regarded as effective tools for supporting time-consuming optimization studies by learning complex relationships between experimental parameters and pollutant removal efficiency [33]. Advanced algorithms such as artificial neural networks (ANN), support vector machines (SVM), and eXtreme Gradient Boosting (XGBoost) have been widely utilized for modeling and prediction in electrooxidation processes [34,35]. For instance, Wen et al. [36] applied the XGBoost model to an anaerobic bioreduction process for azo dye removal and achieved a prediction accuracy of 91.6%.

The effectiveness of electrooxidation processes varies significantly depending on the chemical composition of the solution, particularly factors such as the ionic species present in the matrix, organic matter concentration, and buffering capacity. Because real wastewater contains numerous organic and inorganic components, studies conducted in such complex matrices often make it difficult to accurately analyze the independent effects of electrochemical parameters [37]. Therefore, the use of synthetic dye solutions as an initial step to clarify the electrooxidation mechanism and determine optimal operating conditions is a widely adopted approach in the literature [38,39]. In line with this approach, the current study was designed to determine the effective parameters in the electrooxidative removal of Trypan Blue (TB), an azo dye class, using a BDD anode under controlled conditions before applying it to real wastewater.

Our hypothesis in this study is that integrating the XGBoost algorithm-based modeling with systematic optimization of electrooxidation parameters will provide an approach that accurately predicts dye removal efficiency and identifies more energy-efficient operating conditions. For this purpose, the removal efficiency of TB from aqueous solutions via electrooxidation was investigated. BDD electrodes were used as anodes, and stainless steel was used as the cathode material. A machine learning model based on the XGBoost algorithm was developed to predict and optimize process performance. Previous studies in the literature have generally focused on the experimental optimization of electrooxidation parameters or the application of machine learning methods to other wastewater treatment processes. In contrast, this study presents an integrative approach by integrating detailed experimental analysis of electrooxidation using BDD anodes with XGBoost-based predictive modeling. This two-pronged approach provides a deeper understanding of electrooxidation performance and a powerful tool for predicting dye removal efficiency with high accuracy. Therefore, this study adds new insights to the literature and proposes an innovative framework for optimizing wastewater treatment processes.

2. Materials and Methods

2.1. Experimental Equipment and Materials

In this study, the removal of TB was investigated using the electrooxidation method. A machine learning model based on the XGBoost algorithm was developed to predict and optimize process performance. Electrooxidation experiments were conducted using a 250 mL synthetic solution under continuous stirring, with BDD electrodes (Changsha 3 Better Ultra-Hard Materials Co., Ltd., Changsha, China) as the anode and stainless steel (SS 316) as the cathode. Iron electrodes were also used during the tests. The BDD anode and stainless steel cathode were rectangular plates (55 × 120 mm). The effective geometric surface area was 132 cm². Anhydrous sodium sulfate (Na₂SO₄; M = 142.04 g/mol) was used as the supporting electrolyte at 20–100 mM (2.84–14.20 g/L). The initial electrical conductivity of the solutions prior to electrolysis was 10,950–38,700 µS/cm at 25 °C. The tested current densities are listed in

Table 1. Experimental Conditions.

Parameters	Values
Current Density (mA/cm2)	0.152, 0.378, 0.530, 0.757, 1.136
Stirring Speed (rpm)	200, 400, 600
Initial Dye Concentration (mg/L)	100, 200, 400
pH	2, 5, 6, 8, 11
Supporting electrolyte concentration (Na2SO4, mM)	20, 40, 60, 80, 100

.

The electrooxidation process was performed in a reactor constructed from circular-bottom Plexiglas, ensuring homogeneous mixing of the 250 mL synthetic solution. The molecular formula of TB used in the study is C₃₄H₂₄N₆Na₄O₁₄S₄, with a molecular weight of 960.81 g/mol. Its chemical structure and UV–Vis absorption spectrum are presented in Figure 1. A stock solution of 1000 mg/L TB was prepared for the experiments. A GW INSTEK GPS-4303 Laboratory DC (Suzhou, China) power supply was used to provide direct current.

The pH of the solutions was adjusted using 0.1 N H₂SO₄ (Merck, Darmstadt, Germany, 95–97%) and 0.1 N NaOH (Sigma-Aldrich, Saint Louis, MO, USA). All chemicals used in the study were of analytical grade. After each experiment, pH measurements were performed using a WTW Multi 3620 IDS SET C pH meter (Troistedt, Germany). Absorbance measurements were conducted spectrophotometrically using an Optizen POP UV-Vis spectrophotometer (Mecasys Co., Ltd., Daejeon, Republic of Korea) at a wavelength of λ_max = 590 nm. All experiments were carried out at a constant temperature of 25 °C.

In this study, key operational parameters including pH, current density, dye concentration, and supporting electrolyte concentration were optimized to enhance TB removal efficiency.

All electrooxidation experiments were conducted at a constant temperature of 25 °C to ensure thermal consistency throughout the reaction process. The total reaction time for each run was set at 60 min. The experimental conditions, including current density, stirring speed, and initial dye concentration, are summarized in Table 1. A schematic representation of the experimental setup used in this study is provided in Figure 2.

The percentage degradation of TB was determined using Equation (3).

$$\mathrm{TB} \text{degradation} (\%)={\displaystyle \frac{C_0-C_e}{C_0}}\times 100$$

(3)

where C₀ (mg/L) and C_e (mg/L) represent the initial and final concentrations of TB in solution, respectively.

Energy consumption is a critical parameter for assessing the feasibility and practical applicability of the electrochemical dye removal process [26]. In this study, energy consumption (EC) was expressed in watt-hours per liter (Wh/L) and calculated using Equation (4) [40]:

$$EC={\displaystyle \frac{I\times V\times t}{v}}$$

(4)

In this equation, V is the applied voltage (V), I the applied current (A), t is the electrolysis time (h), and v is the volume of the dye-containing wastewater in the reactor (L).

2.2. Data Analysis Methods

eXtreme Gradient Boosting Algorithm (XGBoost)

This study used the XGBoost algorithm to estimate the azo dye removal efficiency achieved by the electrooxidation process. XGBoost is a machine learning algorithm optimized and improved based on the gradient boosting method [35,41]. Compared to traditional gradient boosting algorithms, the XGBoost algorithm stands out with its high accuracy, short computational time, and mechanisms that prevent overfitting [35]. The XGBoost algorithm improves the overall accuracy performance of the model after an optimization process based on errors. In addition, XGBoost is an ideal learning algorithm for large and complex datasets by offering advanced features such as automatic tree pruning and cross-validation [35]. In this context, in the current study, hyperparameter optimization was performed to increase the prediction performance of the model. For this purpose, the following hyperparameters were optimized by applying the Grid Search method.

-

Learning Rate (eta): It determines the learning rate that the model can perform in each iteration. Smaller values (e.g., 0.01–0.3) allow the model to learn more consistently but at a slower pace [35].

-

Max Depth: It refers to the maximum depth of decision trees. As the depth of the tree increases, the model’s capacity to learn complex relationships increases, while the risk of overfitting also increases [35].

-

Subsample: It helps prevent overfitting by determining the proportion of training data to be used in each iteration [42].

-

Colsample_bytree: It defines the ratio of features to be used for each tree. In addition, this parameter increases the generalization ability by increasing the diversity of the model.

Experimentally obtained parameters of electrooxidation were used to train the model. Operational parameters—pH (2, 5, 6, 8, 11); current density (0.152, 0.378, 0.530, 0.757, 1.136 mA/cm²); electrolysis time (0–60 min; each run lasted 60 min); supporting electrolyte (Na₂SO₄) concentration (20, 40, 60, 80, 100 mM); initial dye concentration (100, 200, 400 mg/L); and stirring speed (200, 400, 600 rpm)—were used as input variables, while TB removal efficiency (%) was used as the output variable. Data were divided into 80% training set and 20% test set. The model was trained on the training set and the prediction performance of the model was evaluated with the test set.

Determination Coefficient (R²), Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) were used to evaluate the accuracy and generalization ability of the model.

The study findings showed that the XGBoost algorithm can predict the azo dye removal efficiency in the electrooxidation process with high accuracy and provides an effective tool for the optimization of the process. In addition, thanks to the XGBoost model, the most efficient parameter combinations of the process were determined and thus the experimental design time was reduced [36].

In this study, R software (version 4.4.1) environment [43] was used for machine learning-based estimation of azo dye removal efficiency by electrooxidation process. Modeling and analysis processes were carried out via RStudio (version 2025.09.0+387) [44] interface. Firstly, using the XGBoost package (version 1.7.8.1) [45], which is the R adaptation of the XGBoost algorithm, tree-based gradient boosting models were established quickly and efficiently; thanks to its parallel processing capability, computation time was reduced in large datasets. In this study, hyperparameter optimization and model training were performed directly through this package. The caret package [46] was used to standardize and simplify the modeling process; model training, cross-validation, hyperparameter adjustments and performance evaluation processes were carried out through this package, and the most appropriate hyperparameter combinations were determined, especially with the Grid Search method. The “dplyr” package was used for data pre-processing to summarize and organize the data [47]. The “ggplot2” package was preferred for visualizing the obtained model results [48]. In this way, the model performances between actual and predicted values were expressed graphically. Finally, the “Metrics” package was used to calculate the model’s performance evaluation metrics (RMSE, MAE, and R²) [49].

3. Results and Discussion

3.1. Effect of the Operational Parameters

3.1.1. Effect of pH on TB Removal

In electrochemical processes, the pH of the solution directly influences the formation, type, and quantity of oxidant species generated in the system, thereby playing a critical role in determining the removal efficiency of organic pollutants. In this study, the effect of initial pH on TB dye removal was investigated, and the results are illustrated in Figure 3.

The experiments were conducted under constant operating conditions, including a supporting electrolyte concentration of 60 mM Na₂SO₄, a temperature of 25 °C, a stirring speed of 200 rpm, and an initial TB concentration of 100 mg/L. The initial pH values tested were 2, 5, 6, 8, and 11.

The results demonstrated that the highest TB removal efficiency was achieved under acidic conditions, particularly at pH 2. In low-pH environments, adsorbed (physisorbed) •OH formed on the non-active BDD anode exhibit exceptionally high oxidation potential, thereby accelerating the oxidative degradation/decolorization of organic pollutants [50]. Due to their inert surface properties, BDD anodes facilitate oxidative degradation through radical mechanisms, establishing a highly oxidative environment within the solution [51]. At low pH values, proton concentration increases, which synergistically enhances the degradation of TB molecules in conjunction with the strong oxidative effect of the anode [52].

Beyond radical oxidation on BDD, acidic pH can alter the state of TB. As a sulfonated diazo dye, TB undergoes protonation-dependent azo–hydrazone tautomerism and tends to aggregate at low pH, which induces halochromic shifts and lowers the apparent molar absorptivity; these speciation/aggregation effects, together with a compressed double layer, facilitate decolorization at pH 2 [53,54,55].

Concomitantly, higher proton/ionic concentration at low pH compresses the interfacial electrical double layer, reducing electrostatic repulsion and bringing anionic dye molecules into closer proximity to the anode surface where •OH is generated, thereby increasing reaction probability [56]. This interaction brings TB molecules into closer proximity to the anode, thereby facilitating more effective contact with surface-generated hydroxyl radicals. This proximity effect is a key contributor to the enhanced removal rates observed at low pH.

In contrast, a significant decline in removal efficiency was observed under neutral and alkaline conditions. At pH 8 and 11, the color removal efficiencies after 60 min of treatment dropped to approximately 60% and 40%, respectively. As the pH increases; however, an excess of hydroxide ions (OH⁻) can inhibit the adsorption of water molecules onto the anode. This inhibition reduces the generation of the electrochemically active •OH species, often leading to a prevalent oxygen evolution reaction that results in the production of molecular oxygen—significantly less effective for degrading chromophore structures of reactive dyes such as that seen with the use of BDD [57]. Under alkaline conditions, the excess of hydroxide ions inhibits the generation of hydroxyl radicals on the anode surface, favoring instead the formation of less reactive oxygen gas. This behavior is characteristic of BDD anodes and restricts the overall oxidative efficiency. Moreover, at high pH, the increased negative charge on TB molecules enhances electrostatic repulsion from the similarly charged anode surface, which further reduces their interaction and degradation.

The use of a sulfate-based supporting electrolyte (Na₂SO₄) can enhance the oxidative capacity of the system by promoting the formation of sulfate radicals (SO₄•⁻). However, the generation and stability of these radicals are also pH-dependent, with optimal activity observed under acidic conditions [58]. At high pH, the conversion of SO₄•⁻ radicals to •OH radicals are limited, resulting in a reduced total oxidative capacity.

Overall, considering the use of BDD anodes and the anionic nature of TB, acidic pH optimizes both oxidant generation and dye–surface interactions, yielding the highest decolorization efficiencies. These findings are consistent with previous studies employing BDD electrodes [58,59], highlighting the critical influence of pH on electrooxidation performance. Furthermore, the results emphasize the importance of determining optimal pH conditions in the design of industrial electrooxidation systems to achieve high removal efficiency and improved energy utilization.

3.1.2. Effect of Mixing Speed on TB Removal

Stirring speed is an important operational parameter in electrooxidation because it modulates external mass transfer—i.e., the transport of reactants across the hydrodynamic/diffusion boundary layer toward the anode surface. In electrooxidation with BDD—a non-active anode—oxidation proceeds predominantly via adsorbed (physisorbed) •OH formed at high oxygen-evolution overpotential; hence, hydrodynamics primarily influences how efficiently TB molecules reach these reactive sites rather than any coagulation/flocculation step [60,61].

In this study, the effect of stirring speed on the removal efficiency of TB was investigated under constant operating conditions: pH 2.0, current density of 0.757 mA/cm², 60 mM Na₂SO₄ as the supporting electrolyte, an initial dye concentration of 100 mg/L, and a temperature of 25 °C. Experiments were performed at three different stirring speeds: 200, 400, and 600 rpm. The experimental results revealed that the highest color removal efficiency—99.73%—was achieved at 200 rpm. At 400 rpm and 600 rpm, the final removal efficiencies were slightly lower, measured as 99.31% and 99.62%, respectively. These results are presented in Figure 4. As shown in the figure, although high removal efficiency was achieved at all stirring speeds, the fastest and most effective decolorization occurred at 200 rpm, where both the initial and final performance were superior.

An increase in stirring speed can enhance the efficiency of oxidation reactions by accelerating the transport of reactant species to the anode surface during the initial stages of electrooxidation. However, at higher stirring speeds, such as 400 rpm and 600 rpm, turbulence generated within the system may adversely affect mass transfer efficiency—particularly by facilitating the accumulation of gas bubbles on the electrode surface. This accumulation can reduce the effective contact time between TB molecules and the anode, thereby limiting overall electrooxidation efficiency [62,63,64].

It is essential to clearly distinguish these electrooxidation-specific trends from electrocoagulation, where mixing regimes are determined to optimize floc formation (rapid mixing followed by gentle flocculation) and excessive shear breaks up the flocs [39,65]. Such coagulation/flocculation mechanisms are not dominant in the BDD-based electrooxidation system [20].

In summary, the findings indicate that the optimum stirring speed is 200 rpm, at which the electrooxidation system exhibits maximum color removal efficiency. These results underscore the importance of careful optimization of stirring speed, as excessive agitation can impair system performance. In industrial-scale applications, determining the optimal stirring speed is crucial not only for maximizing treatment efficiency but also for minimizing energy consumption and ensuring sustainable process operation.

3.1.3. Effect of Current Density on TB Removal Efficiency and Energy Consumption

In electrochemical oxidation processes, current density is one of the fundamental parameters influencing the generation rate of reactive species and, consequently, the efficiency of organic pollutant removal [66]. In this study, the effect of different current densities on TB removal was investigated using a BDD anode under constant conditions: 100 mg/L initial dye concentration, 60 mM Na₂SO₄ as the supporting electrolyte, pH 2.0, a temperature of 25 °C, and a stirring speed of 200 rpm for 60 min. The applied current densities were 0.152, 0.378, 0.530, 0.757, and 1.136 mA/cm². The experimental results are presented in Figure 5 and Figure 6.

The findings revealed a significant increase in color removal efficiency with increasing current density. After 60 min of electrolysis, the color removal rate was 72% at the lowest current density of 0.152 mA/cm². At 0.378 mA/cm² and 0.530 mA/cm², removal efficiencies of 88% and 99% were achieved, respectively. At the higher current densities of 0.757 and 1.136 mA/cm², 99% color removal was maintained (Figure 5). This trend can be attributed to the enhanced generation of •OH on the surface of the BDD anode. At elevated current densities, these radicals interact more rapidly and effectively with the anionic TB molecules, thereby increasing the rate of oxidation [67].

The use of Na₂SO₄ as a supporting electrolyte reduces cell voltage and enhances mass transfer by increasing the electrical conductivity of the solution. Additionally, sulfate ions (SO₄²⁻) may be oxidized at the anode surface to form SO₄•⁻, which are potent oxidizing agents [58]. These species can act synergistically with hydroxyl radicals, particularly under acidic conditions, to promote more efficient degradation of organic compounds. However, the increase in current density also promotes side reactions that consume active oxidant species. At high current densities, a substantial portion of hydroxyl radicals is diverted to parasitic reactions. As shown in Equation (5), •OH radicals can recombine to form molecular oxygen (O₂), while Equation (6) leads to the formation of hydrogen peroxide (H₂O₂) [67,68,69]. Additionally, the oxidation of sulfate ions can lead to the formation of peroxodisulfate ions (S₂O₈²⁻) Equations (7) and (8) and ozone (O₃) Equation (9), which are associated with weaker oxidative activity [69,70,71].

These competing reactions reduce the availability of active oxidant species for target pollutant degradation, thereby limiting the system’s oxidative capacity and ultimately decreasing color removal efficiency.

$$2BDD(•OH)\to 2BDD+O_{2\left(g\right)}+{2H}^{+}+2e^{-}$$

(5)

$$2BDD(•OH)\to 2BDD+H_2{O}_2$$

(6)

$${S{O}_4•}^{-}+•OH\to {S{O}_4}^{2-}+O{H}^{-}$$

(7)

$$2S{O}_4^{2-}\to S_2{O}_8^{2-}+2e^{-}$$

(8)

$$3H_{2}O\to O_{3\left(g\right)}+6H⁺+6e^{-}$$

(9)

An increase in current density led to a notable rise in energy consumption. While energy consumption was measured at 0.85 Wh/L at the lowest current density of 0.152 mA/cm², it increased to 2.65 Wh/L at 1.136 mA/cm² (Figure 6). This increase is attributed to the application of higher voltage and current values, which elevated the total energy demand of the system. Despite the higher energy input, the marginal improvement in color removal efficiency at elevated current densities resulted in reduced energy efficiency.

These findings are in agreement with previous studies reported in the literature [72,73,74]. Martínez-Huitle and Brillas [70] emphasized that the increase in undesired side reactions at high current densities diminishes the energy efficiency of the electrochemical system, and that optimal operating conditions should be carefully determined. Based on the results of this study, current densities of 0.530 mA/cm² and 0.757 mA/cm² were identified as optimal, considering both energy consumption and color removal efficiency.

Optimizing the applied current density is crucial for minimizing the operational costs of the electrooxidation process by reducing energy consumption, thereby contributing to the development of more sustainable and energy-efficient wastewater treatment systems.

In the electrooxidation process, electrode dissolution may occur due to electrochemical corrosion at both anodic and cathodic surfaces. The theoretical electrode consumption can be estimated using Faraday’s law, which relates the electrode mass loss to the current density, time, and electrochemical parameters. Accordingly, in this study, the electrode consumption (C_E, g/L) for the electrooxidation process was expressed using Equation (10), according to Faraday’s law [75]:

$$C_E={\displaystyle \frac{ItM}{nFV}}$$

(10)

In this equation, $I$ is the electric current (A), $t$ is the electrolysis time (s), $M$ is the molar mass of the electrode material (g/mol), $n$ is the number of electrons transferred, $F$ is the Faraday’s constant (96,485 C/mol), and V is solution volume (L).

For the SS 316 cathode used in this study, possible metal dissolution may occur through oxidation of iron, nickel, or chromium components [76], following reactions such as Fe → Fe²⁺ + 2e⁻, Ni → Ni²⁺ + 2e⁻, and Cr → Cr³⁺ + 3e⁻. However, under the neutral Na₂SO₄ medium and the low current density range applied (0.152–1.136 mA/cm²), these anodic and cathodic dissolution reactions are thermodynamically limited, and the corrosion rate is expected to be negligible [77].

On the other hand, the BDD anode exhibits exceptional corrosion resistance because of its high oxygen evolution potential (≈2.3 V vs. SHE) and chemically inert sp³⁻ bonded carbon lattice, preventing dissolution even under high anodic polarization [39,76,78]. The presence of SO₄²⁻ in the electrolyte facilitates the formation of SO₄•⁻ and S₂O₈²⁻, which enhance the oxidation of organic pollutants rather than electrode degradation [79].

Consequently, electrode consumption in the BDD/SS 316 system is minimal and primarily governed by indirect oxidation reactions instead of electrode corrosion, confirming the long-term electrochemical stability and sustainability of the process.

3.1.4. Effect of Initial Dye Concentration on TB Removal

In the electrooxidation process, the initial dye concentration is a key parameter that influences the organic load in the reaction medium and, consequently, the overall removal efficiency of the system. In this study, the effect of varying initial concentrations of TB dye solution (100, 200, and 400 mg/L) on electrooxidation performance was investigated using BDD anode.

The experiments were conducted under fixed operating conditions: 0.757 mA/cm² current density, 60 mM Na₂SO₄ as the supporting electrolyte, pH 2.0, temperature of 25 °C, stirring speed of 200 rpm, and electrolysis duration of 60 min. The results obtained are illustrated in Figure 7.

The experimental findings indicate that an increase in initial dye concentration negatively affects color removal efficiency. At an initial concentration of 100 mg/L, a removal efficiency of 99.73% was achieved after 60 min. However, for initial concentrations of 200 mg/L and 400 mg/L, the corresponding removal efficiencies decreased to 82.3% and 59.15%, respectively.

As shown in Figure 7, this trend can be attributed to the reduced availability of reactive oxidant species per dye molecule at higher concentrations. At lower dye concentrations, active species—particularly •OH—can interact more effectively with the dye molecules, as there is less competition for these oxidants, resulting in higher removal efficiency.

An increase in the initial dye concentration may elevate the organic load in the system, leading to an insufficient availability of reactive oxidant species—particularly •OH generated on the surface of the BDD anode. Under such conditions, a lower ratio of oxidant species to organic compounds results in decreased color removal efficiency as TB concentration increases. In other words, this leads to an imbalance in the oxidant-to-reductant ratio within the reaction medium, hindering the complete mineralization of dye molecules.

The relatively low removal efficiency observed at 400 mg/L can be attributed to the inability of the generated oxidant species to oxidize dye molecules rapidly and completely. Additionally, elevated dye concentrations may promote the formation of a passive layer on the electrode surface, which restricts mass transfer and diminishes the effectiveness of electrode reactions.

The reduction in removal efficiency despite increased dye concentration also implies that the process becomes less energy-efficient when evaluated on a per-unit removal basis. At higher concentrations, the degradation of dye molecules becomes more challenging, requiring longer electrolysis durations or increased energy input to achieve target removal levels.

Consistent with the current findings, previous studies have also reported that color removal efficiency decreases, and energy consumption increases with rising initial dye concentration [26,40]. This phenomenon is primarily attributed to the limited availability of active oxidant species in the system, which are insufficient to cope with the increased organic load.

Based on the results of this study, electrooxidation provides both high removal efficiency and reduced energy consumption at lower initial dye concentrations, highlighting the importance of this parameter in system optimization. This consideration is especially important during the scale-up phase of electrochemical wastewater treatment systems. Moreover, it is critical for improving energy sustainability and reducing operational costs. For the treatment of wastewater with high organic loads, considering the initial concentration is essential to ensure optimal energy efficiency and long-term process stability.

3.1.5. Effect of Support Electrolyte Concentration on TB Removal Efficiency

In electrochemical oxidation processes, the concentration of the supporting electrolyte is a key parameter that enhances solution conductivity, thereby reducing internal resistance losses and facilitating the generation of oxidant species. In this study, the effect of varying concentrations of Na₂SO₄ as the supporting electrolyte on TB removal efficiency was investigated.

The experiments were conducted over a 60 min electrolysis period under fixed conditions: initial dye concentration of 100 mg/L, current density of 0.757 mA/cm², pH 2.0, temperature of 25 °C, and stirring speed of 200 rpm. The Na₂SO₄ concentrations tested were 20, 40, 60, 80, and 100 mM. The results obtained from these experiments are illustrated in Figure 8.

Experimental findings revealed that increasing the concentration of the supporting electrolyte initially enhanced color removal efficiency; however, beyond a certain threshold, a decline in performance was observed. After 60 min of electrolysis, color removal efficiency was recorded as 98% at 20 mM Na₂SO₄, and 99% at both 40 mM and 60 mM concentrations. However, a further increase to 80 mM and 100 mM led to decreased efficiencies of 98% and 96%, respectively.

Increasing the concentration of Na₂SO₄ enhances the solution’s conductivity, reduces ohmic losses, and promotes a more stable electrochemical environment. Nevertheless, at higher concentrations, the excessive oxidation of SO₄•⁻ at the anode surface may lead to the accumulation of intermediate species such as S₂O₈²⁻, which can trigger undesirable side reactions and negatively impact system efficiency. Furthermore, the elevated ionic strength may increase the solution’s viscosity, thereby hindering the diffusion of oxidant species formed at the electrode surface and limiting reaction kinetics [70]. In addition, conductivity measurements were carried out for different Na₂SO₄ concentrations to evaluate the influence of ionic strength on the electrochemical environment. After 60 min of electrolysis, as shown in

Table 2. Variation in solution conductivity with Na₂SO₄ concentration after 60 min of electrolysis.

Na2SO4 Concentration (mM)	Conductivity (µS/cm)
20	10,950
40	19,160
60	26,800
80	31,800
100	38,700

, the conductivity of the solution increased proportionally with the electrolyte concentration, confirming the enhancement of ionic strength at higher Na₂SO₄ levels [80,81]. This increase in conductivity improves electron transfer and reduces ohmic losses; however, beyond a certain threshold, it may promote side reactions and hinder oxidant diffusion due to increased viscosity and mass transfer resistance [39,80,81,82].

In addition, high concentrations of the supporting electrolyte can stimulate the production of secondary oxidants such as ozone and hydrogen peroxide through electrochemical reactions on the BDD anode surface (Reactions 3–7), enhancing the solution’s oxidative potential. However, surpassing the optimal concentration—particularly at 80 mM and 100 mM—may result in increased mass transfer resistance and promote competitive side reactions at active sites, ultimately diminishing the system’s overall oxidative efficiency.

These findings are consistent with previous reports in the literature. Several studies have demonstrated that moderate concentrations of Na₂SO₄ enhance indirect electrooxidation mechanisms, whereas excessive concentrations adversely affect process efficiency [26,71,83,84].

In this study, the optimum concentration range of Na₂SO₄ as the supporting electrolyte for TB removal via electrooxidation was determined to be between 40 mM and 60 mM. Within this range, high color removal efficiency and a stable electrochemical environment were achieved. In contrast, higher concentrations led to a decline in the effectiveness of oxidation processes and overall system performance. Proper selection of supporting electrolyte concentration is thus essential to maximize process efficiency and promote energy savings in industrial electrooxidation applications.

3.2. XGBoost Model Results

This section presents the experimental results of azo dye removal using the electrooxidation process, along with the predictive outcomes obtained via the machine learning-based XGBoost model. Initially, the effects of key operational parameters—including pH, current density, electrolysis time, initial dye concentration, stirring speed, and supporting electrolyte concentration—on dye removal efficiency were assessed based on experimental data. Subsequently, a predictive model was developed and evaluated using the XGBoost algorithm, and its predictions were compared with the experimental results. The model’s performance was analyzed comprehensively to evaluate its accuracy and reliability.

The findings demonstrated that the influential parameters for optimizing the electrooxidation process were successfully identified, and that the XGBoost algorithm exhibited strong potential for accurately predicting process efficiency. Model performance was quantitatively evaluated using statistical indicators such as the coefficient of R², MAE, and RMSE, confirming the model’s robustness and predictive capability.

Figure 9 presents the linear relationships among the experimental parameters used in the electrooxidation process, visualized using correlation coefficients. The correlation matrix illustrates both the direction and strength of associations between variables, with positive correlations represented by blue tones and negative correlations by red tones [85].

According to the findings obtained in Figure 9, a strong and positive correlation was found between removal efficiency and electrolysis time; this shows that prolonging the electrooxidation time increases the removal efficiency of azo dyes [86,87]. In addition, a negative correlation was found between the initial dye concentration and removal efficiency, and it is understood that high dye concentration negatively affects the removal performance [88,89]. The negative correlation between pH value and removal efficiency shows that the electrooxidation process is more effective at lower pH levels. The negative correlation between current density and pH suggests that the medium becomes acidic with increasing current density or that low pH environments may be more effective at higher current densities [86,90]. However, a positive correlation between support electrolyte concentration and stirring speed was observed, indicating experimental conditions where these two parameters are optimized together [90,91]. These correlation findings are important in terms of showing the multivariable structure of the electrooxidation process and which parameters are determinant in process optimization.

In this study, hyperparameter optimization was performed to improve the performance of the XGBoost algorithm and prevent overfitting of the model. Although no previous research has reported the application of hyperparameter tuning in electrooxidation modeling, several studies in other fields have demonstrated that systematic parameter optimization significantly enhances model robustness and predictive reliability [92,93]. The parameters and value ranges used are presented below:

-

nrounds (Boosting Rounds): We tested this hyperparameter between 100 and 500, in increments of 100.

-

eta (Learning Rate): 0.01, 0.05 and 0.1 values were used.

-

max_depth (Maximum Tree Depth): It tested between 3 and 12, in increments of 3.

-

gamma (Minimum Loss Reduction): Values of 0, 0.1 and 0.2 were used.

-

colsample_bytree (Variable Subsampling Rate per Tree): It tested between 0.5 and 0.9, in increments of 0.1.

-

min_child_weight (Minimum Leaf Node Weight): It tested between 1 and 10, in increments of 3.

-

subsample (Sampling Rate): It tested between 0.5 and 0.9, in increments of 0.1.

Table 3. Hyperparameters for the best prediction model.

Hyperparameters
Nrounds	Eta	Max_depth	Gamma	Colsample_bytree	Min_child_weight	Subsample
500	0.1	6	0.2	0.9	4	0.8
Model evaluation criteria
	Train			Test
RMSE	2.101			8.204
MAE	1.528			4.089
R2	0.9966			0.954

presents the hyperparameter settings and model evaluation criteria for the best prediction model created using the XGBoost algorithm. Among the hyperparameters, the nrounds value was set to 500, indicating that the model was trained for 500 iterations. Eta was set to 0.1; this indicates the magnitude of the update made by the model at each step and provides a relatively balanced learning process. Max depth was set to 6, balancing the complexity of the model. The gamma value of 0.2 indicates the minimum loss reduction requirement for branches to occur and plays a role in preventing the model from overlearning. The colsample_bytree ratio was set to 0.9 and the subsample ratio was set to 0.8, providing diversity against the risk of overfitting of the model. In addition, the min_child_weight value was set to 4, preventing the model from over-branching by controlling the minimum observation weight in each branch.

According to the model evaluation criteria in Table 2, the RMSE value was calculated as 2.10 and the MAE value as 1.52 on the training data; this shows that the model provides high accuracy with very low error rates on the training data. The R² was found as 0.9966, indicating that the model explains 99.66% of the variance in the training data. The test data results reveal the generalization performance of the model: The test RMSE value was calculated as 8.20 and the Test MAE value as 4.08. The R² value of 0.9542 obtained on the test data shows that the model has a strong predictive ability on new and unseen data. However, the difference between training and test performance indicates that the model fits the training data better but still shows an acceptably strong performance on the test data [94]. In general, the XGBoost model gives successful results in both the training and test stages with optimized parameters [95].

Figure 10 shows the changes in R², RMSE and MAE performance metrics obtained from the training data of the XGBoost model depending on the eta (learning rate) and max_depth parameters. It is seen that R² values increase as eta and max_depth values increase and the highest accuracy level is provided in the range of eta 0.08–0.1 and max_depth 10–12. Similarly, RMSE and MAE values reach their lowest levels in these parameter ranges, indicating that the model minimizes the error rates in the training data [96]. These findings reveal that the combinations of high tree depth and optimal learning rate increase the explanatory power of the model and improve the prediction accuracy [97].

Figure 11 shows the changes in R², RMSE and MAE values obtained on the test data of the XGBoost model depending on the eta and max_depth parameters. The graphs reveal that the model’s accuracy rate increases, and error values decrease at higher eta (0.08–0.1) and max_depth (10–12) values [98,99]. While the R² value reaches approximately 0.92 levels in these parameter combinations, it is seen that the RMSE and MAE values fall to minimum levels of 11 and 6, respectively [100]. These findings show that the model also exhibits high performance on test data and that the general accuracy and prediction ability are optimized with appropriate parameter combinations [101,102].

Figure 12 presents the scatter plot showing the relationship between the values predicted by the model and the actual observed values. In the graph, the actual values are on the horizontal axis and the values predicted by the model are on the vertical axis. The proximity of the data to the trend line (blue line) reveals the predictive performance of the model. The graph is supported by a regression line representing the linear relationship and the correlation coefficient (r) is calculated as 0.96. This high correlation value shows that the model’s predictions are largely consistent with the actual values and have a strong predictive ability [103]. However, the deviations observed at some points indicate that the model may be less sensitive, especially when approaching extreme values. In general, it is concluded that the XGBoost model produces predictions with high accuracy.

All the findings obtained revealed that the XGBoost algorithm is a highly successful and reliable method in predicting azo dye removal efficiency in the electrooxidation process. The model exhibited a strong performance with high R² values and low error rates (RMSE and MAE) on both training and test data. The best model obtained because of hyperparameter optimization provided high accuracy with the experimental data and produced stable results even in wide parameter ranges. Especially the high correlation coefficient (r = 0.96) obtained in the test data clearly shows the general verification ability and predictive power of the model. The relationships between the parameters and model performance in three-dimensional surface graphics were clearly revealed, and it was confirmed that the optimal parameter combinations significantly increased the model performance [104]. In this context, the XGBoost-based machine learning model makes a significant contribution to the literature as an effective tool in the optimization and efficiency analysis of electrooxidation processes. As a result, the developed model offers a sustainable, fast and highly accurate prediction and optimization approach for azo dye removal.

4. Conclusions

In this study, the effectiveness of the electrooxidation treatment of azo dye-contaminated wastewater using BDD electrodes was evaluated, and a machine learning model based on the XGBoost algorithm was developed to predict process efficiency. Experimental results showed that the electro-oxidation process provides high dye removal efficiency, and system performance is particularly affected by key operating parameters such as current density, pH, electrolysis time, and support electrolyte concentration. Significant increases in dye removal efficiency were observed when the BDD electrode was used, particularly under acidic pH conditions and at optimal current density values.

The developed XGBoost model predicted experimental data with high accuracy, achieving a coefficient of determination (R²) of 99.6% on the training set and 95.4% on the test set. Low prediction errors and hyperparameter optimization increased the model’s generalizability, demonstrating its suitability as a reliable tool for simulating and optimizing electrooxidation performance. Furthermore, model outputs demonstrated that energy consumption and operating costs can be reduced by correctly optimizing process conditions.

These results demonstrate that integrating machine learning approaches—especially the XGBoost algorithm—with electrochemical treatment technologies has significant potential for improving process efficiency and operational control. Optimizing electrooxidation parameters is critical for high pollutant removal, energy savings, and sustainable process management. In this context, the proposed approach offers a promising framework for developing environmentally friendly, advanced oxidation-based technologies for the treatment of industrial wastewater.

In future studies, it is recommended that the developed model be tested on different azo dye types (anionic and cationic dyes, etc.) and real industrial wastewater samples, and that biotoxicity analyses be incorporated into the process. Furthermore, comparing the model with different machine learning algorithms will contribute to determining the scalability and practical applicability of the electrooxidation process.