XGBoost-TabNet Ensemble Model for Prediction of Methylene Blue Adsorption on Activated Carbon Geopolymer Composite

Machine learning (ML) and deep learning (DL) are changing the horizons of the research world by solving complicated problems. In this work, the adsorption of methylene blue by geopolymer-activated carbon composite was studied using ML and DL algorithms. The adsorption data was subjected to 3 different models i.e., TabNet, XGBoost and ensemble TabNet-XGBoost. Three independent variables i.e., pH, MB concentration and contact time, and two dependent variables i.e. % removal and adsorption capacity were used as input data to these models. The effect of individual model and their ensemble model was studied. Moreover, impact of data distribution for training and testing, and hyper parameters optimization were explored. The XGBoost performed better than TabNet in term of lower Mean Absolute Error (MAE), Root Mean Square Error (RMSE), R-squared (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>). whereas Ensemble of TabNet with XGBoost outperformed both models and resulted in further decrease in error and improvement in adjusted R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>. Training and testing data size has significant effect on MAE, RMSE, R2 and best model fitting was recorded at 85% train data set and 15% test data set. The MAE, RMSE, R2 values obtained at 85:15 training: testing data sets were 2.66, 1.49, 0.935 for % removal data and 1.53, 0.99 and 0.954 for adsorption capacity data, respectively. The combination of TabNet and XGBoost effectively applied to both % removal and adsorption capacity.