Software Defect Prediction Model Using AdaBoost based Random Forest Technique

Software testing is a crucial step in the development procedure. Typically, the defects or errors caused by developers are often dealt with a while later in software development, leading to an increased effect or influence of these flaws. In order to stop this from happening, issues must be identified early in the software development process allowing for the effective use of testing resources. As part of the defect prediction process, software modules are categorized as either defect-prone or non-defect-prone. To automate and enhance the prediction of defective software modules a key criterion for the analysis, numerous methods have been anticipated. In this paper, we create a combined model that assesses the presence of defects in a software module. The proposed work has been based on pre-processing, feature dimensionality reduction and classification. The model is tested through a publicly available NASA dataset. We have applied principal component analysis (PCA) for feature dimensionality reduction, which reduces the dimension of the feature vector. During the prediction task, the boosting technique called AdaBoost has been applied to a random forest named ABRF to determine the prediction rate. Several performance metrics which include accuracy, sensitivity, specificity, F1 score and MCC are measured to validate the planned model. The results of the PCA+ABRF method show that the model has an average accuracy of 0.9862 for the KC2 dataset. From the experimental results, it has been observed that the proposed model provides better defect prediction accuracy as compared to other existing models.