Feature Selection Analysis for Quantitative Fault Prediction  
Author Naeem Seliya


Co-Author(s) Taghi M. Khoshgoftaar


Abstract Software defect prediction and analysis is a vital tool for the software project development team, especially the software quality assurance personnel. Software metrics and defects recorded from prior system releases, or similar projects, are used to build defect prediction models that experts can then use to target testing/inspection efforts. The two kinds of defect prediction models are software quality classification models (i.e., faulty or not-faulty) and quantitative defect prediction models (i.e., regression-based). Feature selection has been widely studied in the context of software quality classification models. However, they have been given little-to-no attention for regression-based defect prediction. Feature selection approaches include filter-based rankers or wrapper-based feature subset selectors. With regression-based defect prediction models, wrappers have almost never been studied, while filters have been studied to a certain extent. This paper examines wrapper-based feature selection for a large software system, using three regression-based learners. The overall finding is that wrappers improved performances of two of the three learners for the given case study. However, further studies with other software systems may yield more conclusive results.


Keywords Quantitative Defect Prediction, Software Quality, Software Metrics, Wrappers-Based Feature Selection
    Article #:  24116
Proceedings ISSAT International Conference on Reliability and Quality in Design 2018
August 2-4, 2018 - Toronto, Ontario, Canada