Modeling of the Ni(II) removal from aqueous solutions by ion exchange resin: Comparison of various machine learning approaches

• Five machine learning models were employed to predict Ni(II) removal using a resin. • Effects of the initial Ni(II) concentration, resin dose, pH, and time were assessed. • Pearson correlation showed significant relationships between q e and input variables. • Polynomial model with a degree of 3 demonstrates commendable performance. • SVR and PL models outperformed other models, with SVR showing superior performance. This study aims to investigate the removal of Ni(II) ions from aqueous solutions using an ion exchange resin, focusing on various machine learning approaches to predict the process. The research highlights the efficiency of Amberlite IR120 Na as a strong acidic cation exchange resin, examining its adsorption capacity under varying conditions, including resin dose, initial Ni(II) concentration, solution pH, temperature, and contact time. The adsorption kinetics were accurately described by the pseudo-second-order kinetic model. Additionally, both surface adsorption and intra-particle diffusion played roles in the steps of the adsorption rate. The adsorption isotherm data fitted well with the Langmuir model, indicating a maximum adsorption capacity of 134.8 mg/g. Moreover, machine learning techniques were utilized to predict the resin’s performance, evaluating five diverse models: Support Vector Regression (SVR), Random Forest, Decision Tree, Multi-Layer Perceptron (MLP), and Polynomial Regression. The results showed that the SVR model performed better than the others, with a training R ² of 0.990 and testing R ² of 0.973, along with the lowest mean absolute error and mean squared error. These findings demonstrate the effectiveness of machine learning in accurately modeling the complex relationships within the adsorption process, thus offering valuable insights for optimizing heavy metal removal from wastewater.