Synthetic Data Generators for Advanced Oxidation Processes in Wastewater Treatment Data Gap
Abstract
Scaling Advanced Oxidation Processes (AOP) for wastewater treatment is limited by the lack of large datasets required for effective Machine Learning (ML) optimization of operational parameters such as UV radiation intensity. To address this, this study evaluates multiple synthetic data generation techniques, including SMOTE, Gaussian Copula, TVAE, CTGAN, and a Markov-based method (ALVI), based on predictive accuracy and computational efficiency. Experimental results show that ALVI and SMOTE achieve the best balance between high predictive performance ($(\mathrm{R}^{2}$ up to 0.913) and low computational cost, while CTGAN performs poorly due to instability. These findings demonstrate that synthetic data generation can effectively mitigate data scarcity, enabling more reliable ML-based optimization of UV-based AOP systems and supporting more efficient wastewater treatment.