A new machine learning model for predicting the water quality index

The water quality directly affects the human health and its assessment requires predicting the Water Quality Index (WQI), which is important for supporting the agriculture and industry and protecting the environment. To this end, the present study: 1) introduces a novel hierarchical spatiotemporal graph neural network (HSTGNN) model to overcome the limitations of the multiple linear regression model (MLR), extract the spatiotemporal features, and capture the global and local inter dependencies of the quality parameters, 2) develops a new particle swarm optimization-Nomad algorithm (PSO-NA) to choose the best input scenario, balance the NA’s exploration/exploitation capabilities and improve the accuracy of the PSO and NA, 3) uses the analysis of variance (ANOVA) method to decompose the models’ output uncertainties into those of the parameters and inputs, and 4) compares the HM with CNN-LSTM model (CL), (LSTM)-MLR (SLM), LSTM-MLR (LM), CNN-MLR (CM), SL, LSTM, CNN and MLR model. All these models have been used to predict the WQI of a plain in Iran.