Software Defect Prediction Based on Multivariate Statistical Method and Machine Learning Techniques

An essential objective of software system development is to find and fix defects before schedule delivery that would be expected below numerous circumstances. During software development, software engineers’ activities can result in software faults, causing disappointment in the future. Thus, the prediction of software defects within the initial stages has become an important interest in the field of software engineering. Many software defect prediction methods proposed in the last two decades. Mahalanobis-Taguchi System (MTS) is a multivariate statistical technique commonly used for feature selection and binary classification problems. In this article, the MTS is used for feature selection and defect prediction. This article compares several machine learning techniques on NASA data sets. The results show that MTS perform best with 100% accuracy, precision, recall, F-measure and 0% MAE, RMSE errors.