Modeling the adsorption process for fluoride removal from groundwater by machine learning

Worldwide, groundwater pollution with heavy metals is a severe concern, threatening living organisms and drinking water safety. High fluoride concentration is a common pollutant among various heavy metals found in groundwater. The adsorption method was more convenient, efficient, economically feasible, and eco-friendly for removing the excess fluoride from groundwater. The fluoride removal efficiency depends on the adsorption process variables such as contact time, pH, alumina dose, temperature, and agitation speed. The association between fluoride removal and adsorption process variables is complex and non-linear. The present study developed an artificial neural networks (ANN) model to calculate the effect and analyze the relationship between adsorption process variables and fluoride removal. The ANN model was trained using the backpropagation algorithm. The estimated fluoride removal was in good agreement with the experimental observations, with an accuracy of (R-2 >99.6) for both training and testing datasets, and was superior to the existing models. The accurate predictions exposed that the model could adequately estimate the relationships between adsorption process variables and fluoride removal from groundwater.