Application of machine learning models for zooplankton abundance prediction in ponds of Southeastern Coastal Regions in Bangladesh.

Zooplankton abundance prediction in surface water bodies is crucial because they reflect ecosystem health and have role in aquatic food webs and nutrient cycling. This study examined the applicability of machine learning algorithms to estimate zooplankton abundance in ponds using water quality parameters. Data on zooplankton abundance and water quality parameters were collected monthly. Multiple linear regression (MLR), support vector regression (SVR), random forest regression (RFR), multi-layer perceptron (MLP), and extreme gradient boosting (XGB), were used to observe the relation between zooplankton abundance and the water quality indicators i.e., ammonia (NH3), dissolved oxygen (DO), biochemical oxygen demand (BOD), electrical conductivity (EC), carbon dioxide (CO2), pH, phosphate (PO4), total dissolved solids (TDS), water hardness (WH), water salinity (WS), and water temperature (WT). The findings revealed that the MLP model, outperformed other models during the training and testing stages, achieving the highest accuracy during training (R2 > 0.98) as well as in testing (R2 > 0.88), with increases in R2 by 20.27%, 2.49%, 1.99%, and 0.92% in training, and 23.32%, 3.40%, 12.79%, and 2.23% in testing compared to the MLR, RFR, SVR, and XGB models, respectively. MLP and MLR were the most stable models, while SVR was the least stable. Sensitivity analysis indicated WT as the most influential parameter for predicting zooplankton abundance, followed by DO, BOD, PO4, WS, pH, NH3, CO2, EC, WH, and TDS. The findings have substantial implications for environmental and fisheries management, providing a valuable tool for monitoring aquatic ecosystems.