On the Application of Cross-Project Validation for Predicting Maintainability of Open Source Software using Machine Learning Techniques

Abstract

Design and development of models to predict software maintenance effort is an impending research area as these models help to predict maintenance effort of software system at earlier stages of its development. The predictions of these models help in allocation of limited resources in an optimal way in the test and maintenance phases of software development. Although numeral software maintainability prediction models have been successfully developed in the past using machine learning (ML) and statistical techniques but there is always threat to generalizability of result have prevailed, as these models are validated on the same data set on which they are trained. This study endeavors to improve generalizability of the software maintainability prediction by cross-project validation where prediction model developed on one software project is validated against the other project. To meet our objective we have taken three open source projects written in java language.The performance of the models is evaluated using prevalent the performance measures. Based on the statistical tests; it is quite conclusive that cross project validation can be successfully applied to predict software maintenance effort of open source software.