Prediction of River Water Quality Using Deep Learning Algorithms

Rapid industrialization in Kuala Lumpur and Klang Valley has significantly contributed to the pollution of the Klang River, making regular monitoring essential for maintaining water quality. However, real-world water quality data often contain errors, limiting the effectiveness of traditional prediction methods. This study explores the potential of hybridizing long short-term memory (LSTM) networks with convolutional neural networks (CNN) for forecasting the river water quality index (WQI). The primary objectives are to compare the proposed WQI forecasting models (LSTM and hybridized CNN-LSTM) and assess accuracy improvements through hybridization. By combining CNN’s spatial feature extraction with LSTM’s temporal modeling capabilities, the CNN-LSTM model enhances predictive accuracy and captures complex temporal patterns in water quality data. The hybridized CNN-LSTM demonstrated a higher value of Kling Gupta Efficiency (KGE) and lower prediction errors compared to the standalone LSTM model, showing its capability in following the fluctuating trends of WQI. Overall, these findings helped to further confirm the potential and validate the effectiveness of hybridization in improving the performance of the models. Given the promising performance shown in this study, hybridization of deep learning models can contribute significantly in advancing sustainable water management and environmental monitoring.