Improved Learning from Small Data Sets Through Effective Combination of Machine Learning Tools with VSG Techniques

Abstract

The present study investigates the small data learning problem and proposes the optimum combination of Virtual Sample Generation (VSG) methods and various machine learning tools which would give highest learning accuracy. Although this small data problem has been studied by various researchers in the recent past, the selection of appropriate VSG methods and machine learning tools, which would yield better accuracy, have comparatively received less attention. This study bridges this gap by investigating the learning performance of three popular VSG methods and five well known machine learning tools. The results of the investigation shows that the learning accuracy is dependent jointly on the choice of the VSG method and the machine learning tool. It has been shown that among all the VSG methods, the Trend Similarity Assessment (TSA) method of VSG when combined with Back Propagation Neural Network (BPNN), gives highest learning accuracy.