A Study on Machine Learning Methods' Application for Dye Adsorption Prediction onto Agricultural Waste Activated Carbon

The adsorption of dyes using 39 adsorbents (16 kinds of agro-wastes) were modeled using random forest (RF), decision tree (DT), and gradient boosting (GB) models based on 350 sets of adsorption experimental data. In addition, the correlation between variables and their importance was applied. After comprehensive feature selection analysis, five important variables were selected from nine variables. The RF with the highest accuracy (R2 = 0.9) was selected as the best model for prediction of adsorption capacity of agro-waste using the five selected variables. The results suggested that agro-waste characteristics (pore volume, surface area, agro-waste pH, and particle size) accounted for 50.7% contribution for adsorption efficiency. The pore volume and surface area are the most important influencing variables among the agro-waste characteristics, while the role of particle size was inconspicuous. The accurate ability of the developed models' prediction could significantly reduce experimental screening efforts, such as predicting the dye removal efficiency of agro-waste activated carbon according to agro-waste characteristics. The relative importance of variables could provide a right direction for better treatments of dyes in the real wastewater.