River watch: AI-Enhanced monitoring of vaigai's water quality through IoT and deep learning

Abstract

The sustainable management of river water resources is critical for urban development and ecological health. This study presents a novel hybrid approach utilizing advanced deep learning techniques to monitor and assess the quality and quantity of river water. We established a network of water testing stations along the river, where comprehensive data were collected on physical water levels and contaminant concentrations. Our innovative framework integrates multiple deep learning architectures, including Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), Multilayer Long Short-Term Memory (M-LSTM) networks, alongside ensemble methods like Random Forest and Support Vector Regression (SVR). This combination allows for a robust analysis of complex patterns in the dataset, significantly enhancing detection accuracy, among all models, deep M-LSTM obtained the best results, with an accuracy of 99.67%, Matthews Correlation Coefficient (MCC) of 99.31%, Cohen's Kappa of 99.30%, sensitivity of 99.67%, specificity of 99.43%, and Receiver Operating Characteristic (ROC) score of 100%. Additionally, M-LSTM recorded an F1 score of 98.66%, precision of 98.67%, Mean Absolute Error (MAE) of 0.0133, Mean Squared Error (MSE) of 0.0133, and an R² score of 0.9447. A cloud-based web interface has been developed for real-time monitoring, facilitating efficient data storage and seamless transfer. Our experiments demonstrated an overall accuracy improvement of 4.1% compared to Random Forest (RF) model. The findings underscore the effectiveness of our hybrid approach in providing precise assessments of river water quality, offering valuable tools for stakeholders engaged in water resource management and environmental conservation.