Estimation of melting points of fatty acids using homogeneously hybridized support vector regression

This work develops a hybridized support vector regression (HSVR)-based model for accurate estimation of melting points of fatty acids using their molecular weights and the number of carbon–carbon double bond as descriptors. The development of HSVR-based model is characterized with two stages. The first stage involves training and testing SVR using test-set-cross validation technique with molecular weights and the number of carbon–carbon double bond as descriptors, while the second stage utilizes the estimated melting points obtained from the first stage as descriptor for further training and testing of SVR. The proposed hybrid system therefore demonstrates a better predictive and generalization ability than ordinary SVR. Furthermore, the melting points of sixty-two fatty acids estimated using the proposed HSVR-based model show persistence closeness with the experimental values than the results of other existing predictive models for fatty acids melting points estimation such as Guijie et al. model and Guendouzi model. The developed HSVR-based model is also characterized with higher value of coefficient of correlation and lower value of mean absolute error than that of the existing predictive models. Superiority of the developed HSVR-based model over the existing predictive models in terms of the ease of obtaining its descriptors and the accuracy of its estimates is advantageous to unravel estimation challenges associated with determination of fatty acids melting points.