Enhancing predictive models using principal component analysis and search based metric selection: a comparative study

Abstract

Predictive models are used for the detection of potentially problematic component that decrease product quality. Source code metrics can be used as input features in predictive models; however, there exist numerous structural measures that capture different aspects of size, coupling, cohesion, inheritance and complexity. An important question to answer is which metrics should be used with a predictor. A comparative analysis of metric selection strategies (principal component analysis, a genetic algorithm and the CK metrics set) has been carried out. Initial results indicate that search-based metric selection gives the best predictive performance in identifying Java classes with high cognitive complexity that degrades product maintenance.