Abstract

In this study, an application of automatic relevance determination-based Bayesian neural network (ARDBNN) approach is explored to quantify the influence of water quality variables for groundwater quality modelling of coastal Maharashtra, India. The ARDBNN model is trained using training samples generated from the bounds of water quality parameters. Upon successful training, the trained model is tested in validation and test data sets. Since the value of hyper-parameters in ARD-BNN modelling is important for categorizing the water quality variables, the stability of the hyperparameters is assessed by optimization test history. The groundwater quality map will be useful for management of the groundwater system of coastal Maharashtra. The novel application of this type could be potential alternative scheme for analysing complex/fuzzy, interlinked water quality variables collected around the world