Performance of non-parametric classifiers on highly skewed data

A wide range of non-parametric classifiers have been suggested and developed in the recent years in order to overcome the compulsion of using the classical parametric Maximum Likelihood Classifier (MLC) for non-normal data. The most advanced of these classifiers being the ones based on the Artificial Neural Network (ANN) algorithm, the Support Vector Machines (SVMs) and the Random Forests (RFs). Although a number of researches have established the efficiency of these distribution-free classifiers over the MLC, nothing much has been contributed in to compare the performance of these non-parametric classifiers against each other. With the objective of filling this gap, this study conducts an empirical study to compare the performances of these three machine learning classification algorithms while classifying asymmetric data. RF classifier was found to be best performing among the three classifiers and robust enough to even very high levels of skewness. AMS subject classification: