On the applicability of search-based algorithms for software change prediction

Numerous research studies have claimed that search-based algorithms have the potential to be effectively used in various software engineering domains. An important task in software organizations is to efficiently recognize change prone classes of a software, as it is crucial to plan efficient resource utilization and to take precautionary design measures as early as possible in the software product lifecycle. This assures development of good quality software products at lower costs. The current study attempts to evaluate the capability of search-based algorithms while developing prediction models for identification of the change prone classes in a software. Though previous literature has evaluated the use of statistical category and machine learning category of algorithms in this domain, the suitability of search-based algorithms needs extensive investigation in this area. Furthermore, the study compares the performance of search-based classifiers with statistical and machine learning classifiers, by empirically validating the results on fourteen open source data sets. The results indicate comparable and in some cases even better performance of search based algorithms in comparison to other evaluated categories of algorithms.