Modeling and optimization of oil adsorption capacity on functionalized magnetic nanoparticles using machine learning approach

Abstract

Using magnetic nanoparticles (MNPs) has emerged as a promising solution to capture oil from emulsified oily wastewater due to their high oil adsorption capacity, low toxicity, and reusability. Various factors affect the oil adsorption process using MNPs; thus, optimization of the process is required to achieve higher oil adsorption capacity. Smart models based on artificial intelligence (AI) are becoming popular as advanced computational tools to assess the non-linear relationships of variables in complex processes. In this study, least squares support vector machines (LSSVM) hybridized with the coupled simulated annealing (CSA) algorithm, adaptive network-based fuzzy inference system (ANFIS), and optimization techniques such as gene expression programming (GEP) are used to predict the oil adsorption capacity as a target variable. Oil concentration, mixing time, and MNP dosage as effective parameters are selected as input variables. After conducting experiments, 149 data points are obtained, 80 % of which is used in the training process and the remaining 20 % for the testing step. The performances of the developed models are evaluated using statistical parameters, including the coefficient of determination (R2), mean percentage error (MPE), and mean absolute percentage error (MAPE). According to the results, ANFIS and LSSVM-CSA models have a better performance than the GEP model in estimating the oil adsorption capacity with higher values of R2 (>0.99) and smaller relative errors (close to zero) for all training, testing, and total datasets. Detailed model evaluation and error analysis indicate that the LSSVM-CSA model predicts slightly better than ANFIS with the highest R2 of 0.9921 and a very small MAPE = 3.7597 % over the total dataset. Although the developed GEP model shows an acceptable prediction with R2 > 0.95, the higher distribution of relative errors of the developed model results in a larger MAPE. Moreover, the GEP model computational time is considerably greater than that of the other models. The relative importance analysis using Pearson’s and Spearman’s correlation coefficients indicates that the oil concentration and MNP dosage are the most influential variables that affect the oil adsorption capacity.