Software Fault Prediction using Machine Learning Techniques

Abstract

Fault-prediction techniques are aiming towards prediction of the software modules that are faulty so that it could be beneficial in the upcoming phases of software development. Difference performance criteria are being employed in order to boost the performance of the already existing ways. However, the main issue is the perspective of compiling their performances which is ignored constantly. Classification is the most used tech-nique that is being used for the exclusion of faulty from non-faulty modules. Previous work under this topic has been carried out using different techniques. Here, we have used tried to enhance the process by using feature selection method so as to make the prediction more accurate and less time consuming. Moreover, nine fault prediction techniques have been such as Adaboost, multilayer perceptron, decision tree regression, linear regression, Boosting, random forest, support vector machine, bagging, boosting for the prediction of number of faults. The study has been carried out using ten software project datasets from PROMISE repository. Mainly, AAE, ARE and feature selection techniques have been used to evaluate the results of the investigation. The kruskal-wallis test has been applied on the ARE and AAE values to check whether the difference of performance of the various machine learning techniques is statistically significant or not.