Modeling of BOD and COD Concentrations in Aquaculture Sites of the Southeastern Coastal Region of Bangladesh Using Machine Learning

ABSTRACT Biochemical oxygen demand (BOD) and chemical oxygen demand (COD) are key indicators of aquaculture water quality. However, few studies have explored the applicability of machine learning (ML) approaches for their prediction. Therefore, this study evaluated five ML models using key physicochemical water quality parameters, including water temperature (WT), electrical conductivity (EC), dissolved oxygen (DO), pH, total dissolved solids (TDS), turbidity (T), and water salinity (WS). The multilayer perceptron (MLP) outperformed the other models, improving R 2 by up to 30.92% for BOD and 19.42% for COD compared with the other models. Using MLP, the optimal input combinations (IC‐5), comprising WT, DO, EC, pH, and TDS for BOD and TDS, WS, EC, T, and WT for COD, explained 76.6% of the variance in BOD and 85.6% of the variance in COD, contributing over 93% to the overall prediction accuracy ( R 2 ). These findings may support the practical monitoring, management, and regulation of aquaculture waters.