Module-Order Modeling with Feature Selection  
Author Naeem Seliya

 
Co-Author(s) Taghi M. Khoshgoftaar

 
Abstract Software defect prediction models are commonly used for cost-effective defect detection and correction. A module-order model provides an ordered ranking of program modules based on a quality factor, such as software defects. It uses an underlying regression model to obtain the predicted defects, which are then used to rank the modules. A cutoff percentage is then used to obtain the modules to target for software testing and re-inspection efforts. We present for the first time, to the best of our knowledge, module-order modeling with feature selection. Our case study consists of a real-world Java-based software project. The regression models used are MultilayerPerceptron and Linear Regression, within the WEKA data mining tool framework. A comparison of the models with and without feature selection suggests that the former improves the module-order model performances, for both underlying regression models. However, in both cases the models show poor performance for some of the cutoff values. When comparing both of the regression-based module-order models, i.e., Linear Regression and MultilayerPerceptron, the latter shows better performance for the same cutoff values.

 
Keywords Module-Order Model, Feature Selection, Defect Prediction, Software Metrics, Software Quality
   
    Article #:  DSIS19-35
 
Proceedings of ISSAT International Conference on Data Science & Intelligent Systems
August 1-3, 2019 - Las Vegas, NV, U.S.A.