Defect prediction as a multiobjective optimization problem
Gerardo Canfora, Andrea De Lucia, Massimiliano Di Penta, Rocco Oliveto, Annibale Panichella and Sebastiano Panichella
Article first published online: 8 MAR 2015 | DOI: 10.1002/stvr.1570
When performing defect prediction, a software engineer often wants to achieve conflicting objectives, for example, maximizing the number of identified defect-prone classes and reducing the code inspection cost. In this paper, we formulate the defect-prediction problem as a multi-objective problem and propose an approach called Multi-objective Defect Prediction (MODEP), which uses a genetic algorithm to build defect predictor models. Results of an empirical study indicate that MODEP outperforms single-objective predictors, trivial baselines, and local prediction models.