Software Defect Density Prediction based on Multiple Linear Regression

Abstract

The software quality is an indicator of the ability of a software system to function in accordance with the customer requirements. In maintaining an acceptable software quality level, defect density plays a major role. With the increasing complexity of the software, the chances of defects are also increasing. In this paper, a machine learning-based model for the prediction of the defect density of the future version of an open-source software (OSS) using the multiple linear regression technique is presented. The data of the OSS versions has been collected from the version control system, Git. Software metrics such as source lines of code, number of developers, number of commits, and code churn affect the defect density of the software, therefore, these metrics are selected as the predictor variables. The normality test of each of the variables is performed to show that the variable data belongs to a normal distribution. The results show that how much variability of defect density is predicted by these software metrics.