Identifying Effective Software Metrics for Categorical Defect Prediction Using Structural Equation Modeling

Software Defect prediction is the pre-eminent area of software engineering which has witnessed huge importance over last decades. The identification of defects in the early stages of software development improve the quality of the software system and reduce the effort in maintaining the quality of software product. Many research studies have been conducted to construct the prediction model that considers the CK metrics suite and object oriented software metrics. For the prediction model development, consideration of interaction among the metrics is not a common practice. This paper presents the empirical evaluation in which several software metrics were investigated in order to identify the effective set of the metrics for each defect category which can significantly improve the defect prediction model made for each defect category. For each of the metrics, Pearson correlation coefficient with the number of defect categories were calculated and subsequently stepwise regression model is constructed to predict the reduced set metrics for each defect category. We have proposed a novel approach for modeling the defects using structural equation modeling further which validates our work. Structural models were built for each defect category using structural equation modeling which claims that results are validated.