Navigating viscosity of GO-SiO₂/HPAM composite using response surface methodology and supervised machine learning models

This paper presents the use of response surface methodology (RSM) and robust supervised machine learning approaches to model nano-polymeric viscosity. In the absence of studies in the previous research on using these two techniques, this study's key objective was to equate these strategies' efficiency with a nano-polymer solution's viscosity modelling. Synthesized nano-polymer composite is characterized by scanning electron microscope, transmission electron microscope, X-ray diffraction, and thermogravimetric analysis. The research used the Split-Plot Central Composite Design (SP-CCD) of RSM to model five independent parameters for nano-polymeric composite viscosity, namely shear rate, temperature, and concentration nanoparticles, sodium chloride, and calcium chloride. The findings indicate that the polymer solution's viscosity was not equally crucial to all parameters. Besides, a variance analysis (ANOVA) was conducted, and no evidence of inadequacy in the RSM model was provided. The comparative study of the conventional central composite design model and the supervised machine learning model was conducted. The split-plot central composite model gave R² of 92% and adjusted R² of 88%. Amongst all supervised machine learning models used, the XGBoost model recorded the highest R² accuracy of 90% and lowest prediction errors in terms of mean absolute error and root mean square error. XGBoost when compared to other models in terms of akaike information criterion had the highest likelihood of fit making it the best option in analyzing rheological behavior for hybrid polymeric nanofluid.