An exploratory study for software change prediction in object-oriented systems using hybridized techniques

Abstract

Variation in software requirements, technological upgrade and occurrence of defects necessitate change in software for its effective use. Early detection of those classes of a software which are prone to change is critical for software developers and project managers as it can aid in efficient resource allocation of limited resources. Moreover, change prone classes should be efficiently restructured and designed to prevent introduction of defects. Recently, use of search based techniques and their hybridized counter-parts have been advocated in the field of software engineering predictive modeling as these techniques help in identification of optimal solutions for a specific problem by testing the goodness of a number of possible solutions. In this paper, we propose a novel approach for change prediction using search-based techniques and hybridized techniques. Further, we address the following issues: (i) low repeatability of empirical studies, (ii) less use of statistical tests for comparing the effectiveness of models, and (iii) non-assessment of trade-off between runtime and predictive performance of various techniques. This paper presents an empirical validation of search-based techniques and their hybridized versions, which yields unbiased, accurate and repeatable results. The study analyzes and compares the predictive performance of five search-based, five hybridized techniques and four widely used machine learning techniques and a statistical technique for predicting change prone classes in six application packages of a popular operating system for mobile—Android. The results of the study advocate the use of hybridized techniques for developing models to identify change prone classes.

from #AlexandrosSfakianakis via Alexandros G.Sfakianakis on Inoreader http://ift.tt/2rhb6WD
via IFTTT