Geo-Information Approaches for Water Quality Monitoring in Arid River Systems Using Machine Learning

Water is the most important resource for life; however, the intensifying exploitation of water resources has led to significant degradation in water quality, particularly in rivers. This study investigates the potential of predictive models based on artificial intelligence techniques, such as Support Vector Machine (SVM) combined with mathematical approaches such as the Water Quality Index (WQI), to enhance the forecasting of water quality. We detail the methodology employed to construct these predictive models utilizing SVM, a specific WQI in conjunction with domain expertise. The models are developed from historical physicochemical parameter datasets from seven monitoring stations along a section of the Loa River in Antofagasta, Chile. The performance of the SVM model was rigorously validated using four key metrics: accuracy (acc), precision (p), recall (r), and F1-score. This paper elucidates the processes of dataset curation, and threshold optimization for influential physicochemical parameters. The approach presented herein is innovative, as it marks the first attempt to predict water quality specifically for the Loa River using SVM. The resultant models demonstrate robust performance metrics, achieving mean values of acc = 0.866, p = 0.849, r = 0.863, and F1-score = 0.847, positioning them competitively against analogous studies employing alternative methodologies in similar contexts.