Statistical testing on prediction of software defects

Abstract

Statistical Tests are used to make inferences from data. These tests will tell whether the observed pattern is real or just due to chance. The type of the test, to be used, depends on research design, distribution of data and type of variables. In this paper, we are addressing high dimensionality problem in software defect prediction using statistical tests. We determined the distribution of data to choose appropriate statistical test. We observed most of the variables follow gamma distribution and hence applied wilcoxon Rank Sum Test for correlation between input variables and outcome variable. We extracted the variable with high correlation. We observed the performance of the classifier was improved by addressing high dimensionality problem with wilcoxon Rank Sum Test.