Article Preview
TopIntroduction
The prime objective of any software development model is to ensure that the final product or service has the correct level of quality as per the end user’s requirements, called as Software Quality Assurance (SQA). Any deviation from the actual and expected results for some preset environmental configurations can be described as a defect in terms of end-user specifications. Software Development Life Cycle (SDLC)'s most important stage is testing, as it consumes a large proportion of the project's total cost. Therefore, in every software development cycle this step should be focused first. The only way to address this problem is the Software Defect Prediction (SDP) at the right time.
The traditional approach of Defect Prediction (DP) is to identify "Within-Project" defects by slicing the accessible defect dataset into two subsections so that DP model is trained with one subsection of a dataset (referred to as marked cases) and the other subsection is used to test the designed DP model which means finding marks in target application dataset which are either defective or non- defective for unidentifiable instances (Ambros et al .(2012)). Cross Project Defect Prediction (CPDP) is a research field where software project lacking enough local defect data can use data from other projects to create an effective and efficient defect predictor. Clearly, cross-project information needs to be listed before; to promote CPDP as it is applied locally (Han et al. (2011)). Homogeneous CPDP gathers common software metrics/features from both parent application (DP model is trained using it’s defect data) & target application (DP model is designed for this) (He et al. (2014)). But, in case of HCPDP, there is no requirement of common metrics between datasets of prediction pair. Matched metrics can be found by measuring the correlation coefficient among all possible combinations of software features between two applications. To predict project-wide defects among heterogeneous projects, the combinations of feature pairs showing some uniform kind of variations in their values are taken as common features between considered pair of datasets. In this article, the authors are attempting to forecast defects in software application which have features that are entirely heterogeneous from the feature set of the source application and also depriving of defect data for constructing effective DP model. Figure 1 shows a clear disparity between Homogeneous CPDP & Heterogeneous CPDP.
Figure 1. Classification of Cross Software Projects Defect Prediction
The proposed research work offers a four-phase novel HCPDP model for addressing the same problem. In addition to this, it also focuses on uneven ratio of instances in training dataset for a binary classification problem known as Class Imbalance Problem (CIP). In order to resolve this issue, the proposed work employs resampling techniques to achieve CIL for an imbalanced training dataset. SMOTE is used as OS technique & RUS is used as US technique to tackle skewness in distribution of instances in training dataset. The paper addresses the following key areas of study. In order to boost the accuracy of a SDP model, the article's motivation is to investigate and fix this CIP in imbalanced datasets.
RQ1. Compare the performance of proposed HCPDP model with CIP and after handling CIP using SMOTE & RUS.
RQ2. Which of the two techniques of resampling (SMOTE & RUS) is giving better outcome?
RQ3. Contrast the results of defect prediction for two categories which are WPDP & HCPDP.