Deep Learning for Prediction of Water Quality Index Classification: Tropical Catchment Environmental Assessment

River water quality modeling using crucial artificial intelligent (AI) models has become an essential tool for river assessment and management. The simplified approach of river health assessment involves computation of water quality index (WQI). In this study, WQI calculation is based on six water quality variables (i.e., dissolved oxygen in % saturation (DO% sat), biochemical oxygen demand (BOD), chemical oxygen demand (COD-Cr), pH, suspended solids (SS), and ammoniacal nitrogen (NH3-N)). This study introduces H2O deep learning (DL), random forest (RF), and decision tree (DT) classification models for WQI based on two investigated modeling scenarios-S-I and S-II. S-I uses a Klang River dataset from five monitoring sites representing a "small-scale catchment"; S-II uses a Klang River Basin dataset from 19 stations representing a "large-scale catchment". In both scenarios, H2O DL model demonstrated excellent outcome and RF performed rather better in S-I, in terms of model performance accuracy as well as classification error. In S-I/S-II, the accuracy for H2O DL was 88.6%/87.8%, RF 89%/84%, and DT 83%/77%; moreover, in terms of classification error for S-I/S-II, H2O DL 11.3%/12%, RF 10.8%/16%, and DT 14%/22%. H2O DL showed excellent performance when dealing with nonlinear, non-stationary, and varying number of WQ data, and it was followed by the RF model. This study highlights the application of the novel H2O DL and RF models for prediction of river WQI classes. The classification based on WQI for small catchment is simple, fast, cost-effective, and beneficial for river WQ assessment, management, and policy-making. Nonetheless, for a large catchment hydrological dataset, it is yet to be improved using specific function of AI algorithm.