Application of Machine Learning Models for Prediction of the Sorption Ability of Rubber Chips Filter

In recent decades, the potential danger of heavy metals in effluents and the entry of water into surface and groundwater resources have been increasingly being considered by the international community. The aim of this study is to provide an indirect method for estimating the efficiency of absorption of rubber chips filter for heavy metals lead, zinc and manganese from industrial effluents. Column adsorption test in a pilot system was conducted as a factorial experiment with three factors based on a completely randomized design with three replications. Three factors were studied including particle size (0.5 and 5 cm), filter thickness (10, 30 and 50 cm) and sorbent contact time with solution. The adsorption of the elements was modeled using 6 models of linear regression, regression tree, artificial neural network, random forest, cubist and support vector machine, based on the Field absorption experiments data. The results showed random forest models, artificial neural network, regression tree and cubist had acceptable performance for predicting adsorption efficiency in all three elements. However, according to the R2 coefficient and RMSE, random forest and artificial neural network showed more satisfactory performance than regression tree and cubist model. Evaluation of the importance showed the high importance of the parameter of adsorbent contact time with the metal solution in all machine learning models. The accurate predicted ability of developed models could significantly reduce field experiment workload such as predicting the removal efficiency of rubber chips. The relative importance of variables could provide a right direction for better treatments of heavy metals.