Integrated PCA–RNN approach for surface water quality assessment in the Mahanadi river system

Water quality defines the suitability of water concerning its physical, chemical and biological characteristics. To identify these properties adequately, one needs to comprehend the variation of the significant parameters affecting the water quality of a particular region. The objective of the study is to assess the impact of four alternative feature selection strategies (ANOVA, random forest, mutual information and principal component analysis (PCA)) on improving the prediction accuracy of Water Quality Index (WQI) in the Mahanadi river system. The optimal features are taken as input to the recurrent neural network (RNN) for WQI prediction. There are a total of 22 physicochemical parameters collected from nine monitoring sites. In particular, PCA provided the desirable results that have a major impact on the water quality. The parameters are reduced to 54% in pre-monsoon, 45% in monsoon and 50% in post-monsoon using PCA. The integration of feature selection and deep learning (DL) techniques is thus considered to be a viable option for computing the WQI, resulting in desired performance and a decrease in input parameters. This reduces the expense, effort and time required for monitoring water quality. Moreover, this can be used globally for determining the important parameters of a river.