Machine learning in environmental monitoring: The case of water potability prediction

Ensuring the quality of drinking water is a critical public health issue, especially in the face of increasing industrial pollution and climate change. This study explores the application of machine learning techniques, specifically XGBoost, to predict water potability based on physical and chemical parameters. Initial experiments with a baseline XGBoost model demonstrated moderate success in classification, with particular difficulty in accurately identifying potable water, which is often underrepresented in the dataset.To address this challenge, the SMOTE technique was applied to balance the dataset, resulting in improved recall for the potable water class. Additionally, hyperparameter optimization via RandomizedSearchCV further enhanced the model’s performance, albeit modestly. Reducing the model’s complexity by selecting the most significant features led to a more efficient model with reduced computational overhead, while maintaining comparable accuracy.The findings of this study indicate that an optimized XGBoost model, supported by data preprocessing and feature selection, can be effectively integrated into automated water quality monitoring systems. Such a system would enable real-time detection of water quality deviations, facilitating prompt corrective actions to protect public health. This approach not only demonstrates the applicability of machine learning in environmental monitoring but also highlights the potential for broader application in resource management and public safety.