Abstract
The oil and gas industry creates environmental challenges by generating wastewater contaminated with toxic compounds such as phenol, which requires effective treatment due to its harmful effects on ecosystems. Adsorption, especially using magnetic activated carbon, is recognized as an efficient and economical method for removing these pollutants. In this study, the adsorption behavior of phenol in a fixed column on magnetic activated carbon adsorbent was simulated under various experimental conditions. To accurately predict the phenol adsorption trend, modern machine learning models, including decision trees, support vector regression, AdaBoost, and linear regression, were employed. By analyzing experimental data and simulating nonlinear adsorption processes, these models were able to provide precise predictions of the residual phenol concentration in the adsorption process output under different conditions. The numerical results showed that the AdaBoost model achieved high accuracy with R² values of 0.9943 for training data and 0.9658 for test data, and the lowest absolute errors of 0.0225 for training data and 0.0535 for test data, indicating the model's strong capability in simulating complex nonlinear adsorption processes. Additionally, sensitivity analysis revealed that time, with a value of 0.57, has a major impact on the phenol adsorption process. Initial concentration and flow rate, with 0.139 and 0.12 respectively, also significantly influence phenol adsorption predictions. These results indicate that time and initial concentration have the greatest impact on reducing the residual phenol concentration. Overall, the prediction of residual concentration trends by various AI models closely aligns with actual data, making these models efficient tools for optimizing adsorption processes in various industries. Consequently, AI models serve as powerful tools for analyzing and predicting phenol adsorption behavior, improving industrial and environmental processes.