With the ongoing urbanization and industrialization, the development and application of smart water quality prediction models are of paramount importance for ensuring water quality safety and advancing environmental sustainability. Therefore, this study introduces two types of machine learning methods, namely the Back Propagation Neural Network (BPNN) and the Genetic Algorithm optimized BPNN (GA-BP). The performances of the GA-BP model and the BPNN are then compared in predicting water quality indexes. Here, five different water quality indexes, including dissolved oxygen (DO), ammonium nitrogen (AN), chemical oxygen demand (COD), total phosphorus (TP), and potassium permanganate index (PPI), were selected from the Zhuting section of the Xiang River Basin between July 2017 and September 2022. The findings revealed that the GA-BP model significantly surpassed the BPNN in terms of predictive performance, exhibiting both superior accuracy and robustness. Specifically, in forecasting the PPI within the testing dataset, the GA-BP model achieved an exceptional accuracy rate of 96.4%. Additionally, its strong global search capability prevents the BPNN model from being trapped in local extremes. Therefore, the GA-BP water quality forecast model adeptly integrates the advantages of GA, offering enhanced predictive capability and stability for anticipating future water quality conditions. This method provides robust technical support for water environment management and early warning systems.