An Artificial Immune Based Na?ve Bayes Model For Software Defect Predict

Software defect detection is an important topic in the field of software engineering. Being able to predict the defect of software products before releasing can help to reduce the cost of maintaining the products and improve the quality of the software products. In recent years, a variety of machine learning methods have been applied to software defect prediction work. Naive Bayesian method based on classical mathematical theory. However, the independence assumption based by Naive Bayesian method in practical applications are often difficult to reach. Then the performance of the classifiers will be affected. Therefore, feature selection and feature weighting is particularly important for software defect prediction. In addition, due to class imbalance problem in software predict areas, algorithm with not take into account these three aspects is difficult to achieve the desired effect. The innovation of this paper is as follows:(1)First,based on the existing Naive Bayesian method,this paper proposed a Artificial Immune based Naive Bayesian Method to give the corresponding weights for different features to improve the performance of Naive Bayes classification.(2)Second,taking into account that different kinds of software products depends different feather sets and the dependencies of their feather sets are difference, this paper reference Sequential Backward Selection to remove redundant features. Artificial Immune based Naive Bayes with Recursive Feature Elimination are proposed. It get a feather sorted vector by remove a feather in each iteration. Then the best feather set is got.(3)Finally, we take the class imbalance of software defect into account. This paper presents Semi-supervised of Artificial Immune based Naive Bayes with Recursive Feature Elimination, improves the original over-sampling method with active learning approach. The unmarked "defective sample" are marked during self-training process and added to the training set, so that the defective and non-defective sample are balanced.We compare our proposed methods with some related works on AR software defect databases. This proposed methods have improved to some extent with the compared methods in terms of classify performance and on demonstrate the efficacy of the proposed methods.