Software fault prediction using Mamdani type fuzzy inference system

Abstract

High quality software requires the occurrence of minimum number of failures while software runs. Software fault prediction is the determining whether software modules are prone to fault or not. Identification of the modules or code segments which need detailed testing, editing or, reorganising can be possible with the help of software fault prediction systems. In literature, many studies present models for software fault prediction using some soft computing methods which use training/testing phases. As a result, they require historical data to build models. In this study, to eliminate this drawback, Mamdani type fuzzy inference system (FIS) is applied for the software fault prediction problem. Several FIS models are produced and assessed with ROC-AUC as performance measure. The results achieved are ranging between 0.7138 and 0.7304; they are encouraging us to try FIS with the different software metrics and data to demonstrate general FIS performance on this problem.