AI-Driven Prediction of Dye Degradation in Water for Sustainable Water Quality Assessment

Abstract

Water, an indispensable component of our daily lives, plays a pivotal role in sustaining life on Earth. The importance of water extends not only to basic survival but also to various aspects of human activities. However, its quality is increasingly compromised by industrial pollutants, particularly dye effluents from textile industries, which pose significant threats to aquatic ecosystems and public health. Conventional methods for monitoring water quality and measuring dye concentration are often labour-intensive and time-consuming, limiting their applicability for large-scale assessments. To address this challenge, this study investigates the application of artificial neural networks (ANNs) as a predictive tool for evaluating dye degradation in water, thereby supporting real-time monitoring of water quality. Experimental datasets involving three dye types—monoazo, diazo, and polyazo—are utilized, and preprocessing techniques such as Min–Max and Z-score normalization are applied to enhance model training efficiency. The proposed ANN architecture comprises an input layer corresponding to experimental parameters, one or more hidden layers activated by functions such as sigmoid, rectified linear unit (ReLU), and exponential linear Unit (ELU), and an output layer that predicts the percentage of dye degradation. Model training is performed using the backpropagation algorithm. Among the tested configurations, the combination of Min−Max normalization with ReLU activation achieves the highest performance, attaining an R² value of 0.98 in 50 epochs. These findings demonstrate the model’s effectiveness assessment and predicting dye degradation, offering a practical AI-based approach for rapid and scalable water quality assessment, and contributing to more sustainable environmental monitoring solutions.