Prediction of solubility of hydrogen (H₂) in hydrocarbons using QSPR method: MLR data-driven as a simple Machine Learning (ML) algorithm

In this study, the ‘Quantitative Structure-Property Relationship’ (QSPR) method has been applied for the prediction hydrogen (H₂) solubility in different types of hydrocarbons using a new bigger dataset than former studied datasets. The dataset constitutes of 1751 datapoints including 32 unique hydrocarbons at the wide ranges of pressures and temperatures. The simple Machine Learning (ML) algorithm, called ‘Multilinear Regression (MLR)’ has been applied for the model development for the first time which has not been studied for this application, yet. The two suggested MLR-QSPR models including novel molecular descriptors, called ‘PaDEL’ and ‘sigma profile’ descriptors, have been developed for the first time. The dataset was divided to a training set for the development of models, and to a validation set for external validation. The advantages of this study were discussed and compared with other available models which were developed with other ML algorithms. In these comparisons, some deficiencies of former models have been shown and discussed. Unlike former models, internal validation using Leave One/Multi Out- Cross Validations (LOO-CV/LMO-CV) and Y-scrambling methods were performed on the both MLR-QSPR models using statistical parameters for further assessment. According to the obtained results of statistical parameters (R² = 0.98 and Q²_LOO-CV = 0.98), the predictive capability of the suggested MLR-QSPR models was acceptable for training set. Regarding the external validation, another statistical parameter like AARD% = 9.79 was also satisfactory for validation set.