Building a Novel GP-Based Software Quality Classifier Using Multiple Validation Datasets

One problem associated with software quality classification (SQC) modeling is that the historical metric dataset obtained from a single software project are often not adequate to build robust and accurate models. To address this issue, multiple datasets obtained from different software projects are used for SQC modeling in recent research works. Our previous study has demonstrated that using multiple datasets for validation can achieve robust genetic programming (GP)-based SQC models. This paper further investigates the effectiveness of using multiple validation datasets. Moreover, a novel GP-based classifier consisting of training, multiple-dataset validation, and voting phases, is proposed. The experiments are carried out on seven NASA software projects. The results are compared with the results achieved by seventeen other data mining techniques. The comparisons demonstrate that the performance of our approach is significantly better by using multiple datasets from different software projects with similar reliability goals.