Regional scale modelling for the prediction of arsenic in groundwater in the alluvial plains of Ganga River basin.

Groundwater is an essential source of potable water. Significant public health issues have been documented to arise from the presence of high levels of arsenic (As) around the world, particularly in South Asia, which includes India. Numerous studies have provided evidence for the impact of human and geomorphologic factors on the mobilization and distribution patterns of arsenic in the Ganga River delta. Groundwater samples from Begusarai district, Bihar, India, were collected and tested for various parameters that were used later as input variables in modeling. This study shows the groundwater arsenic prediction using artificial intelligence (AI) [i.e., machine learning] methods, specifically, the Linear regression model (LRM), Random forest regressor model (RFRM), Decision tree regressor model (DTRM), and artificial neural network model (ANNM) for the prediction of arsenic concentration in groundwater (data split: 70% training and 30% testing). Model assessment metrics were determined to validate the performance of applied models. LRM and ANNM have the lowest MSE and RMSE values among the models. LRM has the highest NSE value of 0.793, followed closely by the ANNM with an NSE of 0.781. LRM again shhows the highest R2 value of 0.793, demonstrating the model's improved suitability for prediction of arsenic concentration in groundwater in the study area. [ABSTRACT FROM AUTHOR]