Ensemble of feature selectors for software fault localization

Fault localization is an important process in software development and maintenance. Recently, code coverage information coupled with machine learning techniques (especially filter-based feature selection) have been used to isolate potentially faulty regions of code. In this initial exploratory paper, we propose a novel technique that uses strengths of different types of feature selectors. Here, we mainly focus on an ensemble of two classes of feature selectors: 1) convex feature selectors - that use the underlying properties of convexity of operators, and 2) similarity measures - that are typically non-convex operators. We evaluate the effectiveness of our proposed technique of fault localization by using publicly available programs.