Optimization using artificial intelligence methods is an effective approach to improving the performance of systems and processes. These methods enable the identification of more effective parameters and their optimization to enhance phosphate adsorption efficiency.

The present study focused on developing predictive machine learning algorithms with a dimensionality reduction approach. To develop the predictive model, experimental data on phosphate adsorption by sugarcane bagasse hydrochar were obtained through laboratory-scale adsorption experiments. Five independent input variables, including initial pollutant concentration, contact time, adsorbent mass, solution temperature, and pH, were considered in the training process. Additionally, phosphate adsorption efficiency was considered as the output. Among the applied algorithms, the Extra Trees Regressor (ET) demonstrated relatively better performance in predicting phosphate adsorption efficiency, with an R² value of 0.922, as well as lower RMSE (0.074) and MAE (0.048) values. Based on the results, the two input factors with the greatest impact on phosphate adsorption effectiveness were contact time and initial phosphate concentration. Furthermore, the adsorbent mass was identified as the parameter with the least impact.