Machine Learning-Driven Water Safety Risk Index for Rural Drinking Water Quality Assessment

Access to safe drinking water remains a major public health challenge in many rural communities in developing countries, including Nigeria. This study proposes a machine learning–based framework for rural water quality assessment and the development of a Water Safety Risk Index (WSRI) to support evidence-based decision-making. Physico-chemical, microbial, environmental, and spatial parameters were collected from 150 water samples across selected rural communities in Akwa Ibom State, Nigeria. Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN) models were trained to classify water samples into WSRI-based risk categories and evaluated using accuracy, precision, recall, F1-score, and Receiver Operating Characteristic-Area Under the Curve (ROC-AUC) metrics. RF achieved the highest performance, with an accuracy of 0.822 and a ROC–AUC of 0.928, and was therefore selected for feature importance extraction and WSRI construction. The results identify microbial indicators, turbidity, and proximity to waste disposal sites as the dominant contributors to water safety risk. The proposed WSRI provides an interpretable and scalable tool for classifying water safety risk and supporting early warning systems, targeted interventions, and community-level water management in resource-constrained rural settings.