Prediction of liquid-crystalline property using support vector machine classification

A quantitative structure-property relationships analysis applied on a large dataset of aromatic organic compounds was performed using a methodology based on support vector machine (SVM). The liquid-crystalline property was correlated with a series of structural and geometric molecular descriptors: length of the rigid and flexible sections, molecular mass, and an asymmetry parameter that quantifies the ratio between the total length and molecule diameter. The SVM classification was achieved, analyzing the influence of characteristic elements of this tool: type of kernel and parameters of the method. A comparison with other techniques (eager learners and lazy learners algorithms and neural networks) was made using the same dataset and working conditions. The accurate results provided by the SVM method, the small number of parameters, and ease of handling recommend the SVM tool as an appropriate technique for classification. Copyright (c) 2013 John Wiley & Sons, Ltd.