Software Defect Detection by using Data Mining based Fuzzy Logic

Abstract

Software updates and maintenance costs can be reduced by a successful quality control process. Defect prediction is particularly important during software quality control, and a number of methods have been applied to identify defects in a software system. Quality control studies are based on quality metrics and static code metrics, and each research uses different set of metrics during the process. However, it is uncertain which metric is more significant in a particular study. In this study, NASA software quality dataset is used, and the most significant metric in the dataset is determined using MANOVA. A data mining based fuzzy logic model is developed using the reduced dataset. Gini decision tree is used as the data mining algorithm. Results of ROC analysis showed that the hybrid data mining-fuzzy model produces successful results during defect detection in software quality.