Comparative study of preprocessing techniques for real-time prediction of water quality index

Water quality is a fundamental component in human health, environmental sustainability, and industrial utilization. However, conventional methods in monitoring water quality are slow and cannot operate in real-time. This issue highlights the need for an efficient forecasting approach for water quality assessment. This research aims to explore how a machine learning-based framework can be developed to predict the Water Quality Index (WQI) based on eleven key indicators of water quality. Six regression algorithms were compared in this study which are Support Vector Machine, Multi-Layer Perceptron, Decision Tree, Random Forest, XGBoost, and Gradient Boosting. To ensure model reliability, comprehensive data preprocessing steps were implemented. This involved addressing missing values, outlier elimination, and normalizing data to prevent poor model performance caused by variable imbalance or data distortion. In addition, feature selection and Principal Component Analysis were employed to enhance optimal input features. The models were evaluated using R Squared (R²), Root Mean Squared Error (RMSE) and training time. Among all models, Gradient Boosting with normalized data preprocessing achieved the highest overall performance with an R² value of 0.9943 and an RMSE of just 0.5457. The findings demonstrated that integrating machine learning with effective preprocessing substantially improves the accuracy of WQI prediction. Consequently, this approach can provide a robust foundation for developing reliable and scalable real-time water quality monitoring systems.