Application of Machine Learning Algorithms for the Prediction of Metformin Removal with Hydroxyl Radical-Based Photochemical Oxidation and Optimization of Process Parameters

Abstract

This study investigated the effectiveness of hydroxyl radical-based photochemical oxidation processes on metformin (METF) removal, and the experimental data were modeled by machine learning (ML) algorithms. Hydrogen peroxide (HP), sodium percarbonate (PC), and peracetic acid (PAA) were used as hydroxyl radicals sources. Modeling was conducted using ML algorithms with the integration of additional experiments. Under optimum conditions (UV/PC: pH 5, PC 6 mM, UV/HP: pH 3, HP 6 mM, UV/PAA: pH 9, PAA 6 mM), the METF removal efficiency was 74.1%, 40.7%, and 47.9% with UV/PC, UV/HP, and UV/PAA, respectively. The scavenging experiments revealed that hydroxyl and singlet oxygen radicals were dominant in UV/PC and hydroxyl radicals were predominant in UV/HP and UV/PAA. Nitrate negatively affected UV/HP, UV/PC, and UV/PAA, whereas chlorine had a positive impact. The EE/O were 0.682, 1.75, and 1.41 kWh/L for UV/PC, UV/HP, and UV/PAA, respectively. The experimental results were successfully modeled by ML models with high R² values and low MAE and RMSE values. XGBoost models effectively represent data with generalization by avoiding overfitting. Using ML algorithms to model hydroxyl radical-based photochemical oxidation processes is considered an effective and practical method for future research.