An empirical study of software entropy based bug prediction using machine learning

There are many approaches for predicting bugs in software systems. A popular approach for bug prediction is using entropy of changes as proposed by Hassan (2009). This paper uses the metrics derived using entropy of changes to compare five machine learning techniques, namely Gene Expression Programming (GEP), General Regression Neural Network, Locally Weighted Regression, Support Vector Regression (SVR) and Least Median Square Regression for predicting bugs. Four software subsystems: mozilla/layout/generic, mozilla/layout/forms, apache/httpd/modules/ssl and apache/httpd/modules/mappers are used for the validation purpose. The data extraction for the validation purpose is automated by developing an algorithm that employs web scraping and regular expressions. The study suggests GEP and SVR as stable regression techniques for bug prediction using entropy of changes. © 2016, The Society for Reliability Engineering, Quality and Operations Management (SREQOM), India and The Division of Operation and Maintenance, Lulea University of Technology, Sweden.