Comparative Analysis of Machine Learning Algorithms for Water Quality Prediction

This study aims to identify the influential parameters and heavy metals in water and assess the water quality classification at the Alpine glacial lakes and rivers in three districts of Pakistan. For this purpose, nine water quality parameters (Cd, Cr, Pb, Ni, Fe, As, and TDS) in mg/L, pH, Ec µS/Cm are used to compute the Water Quality Index (WQI). The Boruta approach was utilized for the identification of influential parameters associated with the water quality classes. Moreover, we employed supervised machine learning models, including a decision tree, the k-nearest neighbor method, a neural network model (multi-layer perception), a support vector machine, and a random forest, to predict and validate the water quality class. The performance of all algorithms is assessed by an accuracy metric. The accuracy rates for the validation set were observed to be 83% for the decision tree model, 75% for the K-nearest neighbor method, 83% for the neural network, 88% for the support vector machine, and 88% for the random forest model. Water quality assessments for observed locations specify significant insights, revealing that 49% of the locations exhibit low water quality. According to the current study, the government should address problems with water quality in Pakistan’s impacted areas by implementing suitable measures designed water monitoring systems and innovative technologies.