MULTIPLE LINEAR REGRESSION-ARTIFICIAL NEURAL NETWORK HYBRID MODEL PREDICTING ACENTRIC FACTOR OF PURE ORGANIC COMPOUND

摘要

PURPOSE: A multiple linear regression-artificial neural network(MLR-ANN) hybrid model for predicting the acentric factors of pure organic compounds is provided to improve the performance of prediction. CONSTITUTION: Molecular descriptors for the acentric factors of hydrocarbon-based compounds are prepared. Experimental data is classified based on a training set and a testing set. The optimal MLR model(MLRM) for the training set is searched. Entire samples are divided into three sets, and the optimal ANN model(ANNM) is searched. If the absolute value of the predicted acentric factor difference based on the optimal MLRM and the optimal ANNM is more than an over-fitting preventive reference value, the predicted acentric factors based on the MLRM is adopted as the acentric factors.