An Arithmetic Optimization Algorithm–Support Vector Regression Approach for Predicting Drug Solubility in Supercritical Carbon Dioxide

Abstract

An efficient Arithmetic Optimization Algorithm (AOA) was performed to refine the three hyper-parameters of a support vector regression algorithm (SVR). The outcome approach, namely Arithmetic Optimization Algorithm–Support Vector Regression Approach (AOA-SVR) was applied to predict the solubility of 168 drug compounds in supercritical carbon dioxide (SC-CO2), representing a dataset of 13 inputs, 1 output, and 4490 experimental data points (EDP). 4330 EDP were used to build the model, while 160 data points were hidden as an external test. The optimized model was statistically validated with an average absolute relative deviation (AARD%) of 0.7383%, root-mean-square error (RMSE) of 0.1958, coefficient of correlation (r) of 0.9971, coefficient of determination (R²) of 0.9942, robustness (Q²) of 0.9942, and an akaike’s information criteria (AIC) of -1.1290e+04). The overall results proved good predictive ability and robustness.