A Gamma Test Driven Framework for Predicting Fluoride Adsorption Capacity of Biochar Using Multiple Artificial Intelligence Models

Adsorption of fluoride using biochar signifies a promising and innovative method under current study. The complexity of determining biochar’s adsorption properties cannot be fully depicted by traditional mathematical modelling alone. This study presents a novel contribution by applying multiple artificial intelligence techniques to model fluoride adsorption. By considering both proximate and elemental properties of biochar as input variables, this research utilizes the Gamma test (GT) to identify ideal input combinations for the M5 Tree, SVM (Support Vector Machine), MARS (Multivariate Adaptive Regression Splines), MGGP (Multi-Gene Genetic Programming), and MM-ANN (Modular Multilayer Artificial Neural Networks) models. The comparative analysis of these predictive models was evaluated using various statistical metrics (Nash-Sutcliffe efficiency (NSE), Willmott’s index of agreement (WI), root mean squared error (RMSE), mean absolute percentage error (MAPE), and Legates and McCabe’s index (LM)). The MARS-4 model with four input variables exhibited the highest accuracy with MAPE = 102%, RMSE = 2.56 mg g^-1, LM = 0.57, WI = 0.93, and NSE = 0.83 in predicting fluoride adsorption during the testing phase. The MARS-4 model was followed by M5Tree-4 model. These results highlight the MARS-4 model’s effectiveness and its potential to offer valuable insights for optimizing biochar for fluoride removal. The study showed enhanced predictive accuracy for fluoride adsorption and guidance for future biochar engineering.