An advanced deep learning model for predicting water quality index.

• Importance of predicting Water Quality Index (WQI) for assessing water bodies' health and safety. • Development of hybrid model (CNN-CRNN-M5T) combining CNN, Clockwork RNN, and M5 Tree for WQI prediction. • Enhancement of M5T model's data analysis capability for intricate patterns in water quality parameters. • Utilization of GLM-ANOVA to determine significant input variables for WQI prediction. • CNN-CRNN-M5T model's superiority in reducing MAE and increasing efficiency for spatial and temporal WQI predictions in Malaysia. Predicting a water quality index (WQI) is important because it serves as an important metric for assessing the overall health and safety of water bodies. Our paper develops a new hybrid model for predicting the WQI. The study uses a combination of a convolutional neural network (CNN), clockwork recurrent neural network (Clockwork RNN), and M5 Tree (CNN-CRNN-M5T) to predict a WQI. The M5T model lacks advanced operators for extracting meaningful data from water quality parameters, so the new model enhances its ability to analyze intricate patterns. The general linear model analysis of variance (GLM-ANOVA) is an improved version of the ANOVA. Our study uses the GLM-ANOVA to determine significant inputs. As all input variables had p < 0.050, they were defined as significant variables. Results showed that NH-NL and PH had the highest and lowest impact, respectively. Our study used the CNN-CRNN-M5T, CNN-CRNN, CRNN-M5T, CNN-M5T, CRNN, CNN, and M5T models to predict the WQI of a large basin in Malaysia. The CNN-CRNN decreased testing mean absolute error (MAE) of the CRNN, CNN, and M5T models by 2.1 %, 12 %, and 15 %, respectively. The CNN-CRNN-M5T model increased Nash–Sutcliffe efficiency coefficient of the other models by 4–20 % and 2.1–19 %, respectively. The CNN-CRNN-M5T model was a reliable tool for spatial and temporal predictions of WQI. [ABSTRACT FROM AUTHOR]