Neural network-based prediction of water contamination dynamics in a rapidly developing tropical region

Water contamination by toxic elements, including arsenic (As), iron (Fe), phosphate (PO4), and ammonia nitrogen (NH3–N), represents a critical threat to aquatic ecosystems and human health, especially in fast-growing tropical regions. Reliable prediction of contaminant behaviour is essential for early warning and sustainable water management. This study applies an integrated forecasting framework that combines Artificial Neural Networks (ANN) with time-series analysis to evaluate water quality parameters. The ANN models showed moderate predictive performance, with EC) achieving 50.76% accuracy and chloride (Cl) predictions generating MAE and RMSE values of 37.75 and 41.27, respectively. However, the high MAPE (218.88%) for Cl highlights the difficulty of modelling episodic pollution spikes. Correlation analysis further indicated a strong EC–Cl association (r = 0.99) and generally weak interrelationships among nutrients and metals (|r| ≤ 0.46). Time-series decomposition further revealed strong seasonal patterns, higher contamination during dry months due to limited dilution, and an overall rising trend linked to increasing anthropogenic pressures. These findings underscore the region's seasonal vulnerability and demonstrate that integrating ANN with temporal analysis can enhance early warning systems, improve monitoring strategies, and support proactive water quality protection in fast-developing tropical basins.