Optimization-driven modelling of hydrochar derived from fruit waste for adsorption performance evaluation using response surface methodology and machine learning

Abstract

This study aims to explore the potential of integrating Design of Expert (DOE) with Machine Learning (ML) to optimize and predict the adsorption process of solid adsorbent The prediction and optimization of adsorption performance can be improvised using statistical analysis and advanced predictive tools, resulting in substantial cost and energy savings. Firstly, the Response Surface Methodology-Central Composite Design (RSM-CCD) model was used to design and optimize the experiments on the adsorption of cationic dye using biomass-hydro char. Secondly, Random Forest (RF) was used to train the experimental results of RSM-CCD. It is well-suited for small datasets, withstands noise, and effectively reduces overfitting to predict adsorption performance. RF model demonstrated excellent accuracy, achieving a removal efficacy of 97.4 % with a significant R2 value of 0.9981 compared to the RSM-CCD, which had a removal efficiency of 95.6 % and R2 0.9372. The physicochemical analysis also shows the novel hybrid hydrochar from fruit waste exhibits remarkable characteristics, including a higher content of carbon (78 %) and a surface area of 670 m2/g. In summary, RSM-CCD with ML provides precise optimization and predictions of the adsorption efficacy of the novel hydrochar. This has significant value for industrial applications in the field of material discovery.