Software Defect Prediction using Adaptive Neuro Fuzzy Inference System

摘要

Software Defect Prediction is a major challenge in Software Development process, to reduce the cost of software implementation. Predicting Defective prone modules in software industry greatly reduces the software development cost. Most of the researchers applied various data mining techniques like Adaboost, Neural networks, Random Forest and support vector machines for software defect prediction datasets downloaded from NASA repositories. These datasets are imbalanced in nature. In this paper software defects are predicted using Adaptive Neuro Fuzzy Inference System (ANFIS). Initial Fuzzy Inference System (FIS) was derived using Subtractive Clustering method and then FIS was trained using hybrid learning rule. The performance of the classifier is measured in terms of AuC values for these imbalanced datasets. We compared the results of ANFIS with cost sensitive neural networks. The Receiver operating characteristics (ROC) curves are generated and presented in Result section. The ROC values of ANFIS are found satisfactory compared to cost sensitive Neural networks.