The application of data-driven modelling for the water quality index: a case study in Canada

ABSTRACT The water quality index (WQI) is widely used to assess the overall quality of water resources using numerical values. It is a critical tool for both decision-makers and the public to understand the status of water quality. Many WQIs can be found in the literature under different jurisdictions. However, no site-specific index can be found in Saskatchewan. The current research explores the application of data-driven methods for WQIs in the North Saskatchewan River. In total, 444 samples were analysed using 8 key water quality parameters, over 5 river cross-sections from 2012 to 2022. The National Sanitation Foundation (NSF) index was used as a benchmark. The dissolved oxygen (DO), pH, temperature (T), and turbidity (Tr) were identified as pivotal parameters, through correlation-based feature selection, to reduce input dimensionality and improve model efficiency. Five algorithms were applied, namely M5, particle swarm optimization (PSO), differential evolution (DE), gene expression programming (GEP), and multivariate adaptive regression splines (MARS). Sensitivity analysis was conducted to highlight the influence of DO and pH using M5 and GEP models. The findings underscore the potential of data-driven methods to simplify WQIs, offering a practical tool for informed decision-making.