How Far Does the Predictive Decision Impact the Software Project? The Cost, Service Time, and Failure Analysis from a Cross-Project Defect Prediction Model
share
Context: Cross-project defect prediction (CPDP) models are being developed to optimize the testing resources. Objectives: Proposing an ensemble classification framework for CPDP as many existing models are lacking with better performances and analysing the main objectives of CPDP from the outcomes of the proposed classification framework. Method: For the classification task, we propose a bootstrap aggregation based hybrid-inducer ensemble learning (HIEL) technique that uses probabilistic weighted majority voting (PWMV) strategy. To know the impact of HIEL on the software project, we propose three project-specific performance measures such as percent of perfect cleans (PPC), percent of non-perfect cleans (PNPC), and false omission rate (FOR) from the predictions to calculate the amount of saved cost, remaining service time, and percent of the failures in the target project. Results: On many target projects from PROMISE, NASA, and AEEEM repositories, the proposed model outperformed recent works such as TDS, TCA+, HYDRA, TPTL, and CODEP in terms of F-measure. In terms of AUC, the TCA+ and HYDRA models stand as strong competitors to the HIEL model. Conclusion: For better predictions, we recommend ensemble learning approaches for the CPDP models. And, to estimate the benefits from the CPDP models, we recommend the above project-specific performance measures.
READ FULL TEXT