Prediction of organophosphorus pesticide adsorption by biochar using ensemble learning algorithms.

Abstract

Machine learning (ML) models have become a potent tool for advancing environmentally conscious research in materials science, allowing the prediction of wastewater treatment efficacy using eco-materials. In this study, we showcase the potential of an advanced decision tree-based ensemble learning algorithm to model the eviction of emerging organophosphate-based pesticidal pollutants in aqueous systems. The model is trained using laboratory-based biochar adsorption data, and it establishes the relationship between independent experimental factors and the % organophosphate pesticide adsorption efficiency as the output parameter. We classified the experimental dataset into input and output parameters to build the model. The input parameters included pyrolysis temperature, solution pH, surface area, pore volume, and initial pesticide concentration. Grid search optimization in Python was employed to train the model using sets of input-output patterns. The results indicated that the XGBoost-based ensemble ML framework provides the best forecast for pesticide adsorption on the biochar matrix, achieving high scores for the regularization coefficient (R2train = 0.998, R2test = 0.981). The concentration of the organophosphorus compound and the morphology of biochar significantly influenced the pesticide adsorption behavior. These findings demonstrate the potential of using ensemble learning algorithms for the balanced design of carbon-enriched biomaterials to remove emerging micropollutants from water effectively.