Software fault prediction using error probabilities and machine learning approaches

Abstract

Software fault prediction is used to improve the testing efficiency and software quality by earlier identification of software faults associated with software. The identification of faults is usually carried out using the task of classification. The task of classification utilises the code attributes and other features to predict the fault instances. The detection of software faults is prominently affected by a poor classification decision and hence an improved decision-making model is required to predict the patterns using the attributes collected out from the datasets. In the first part of the research, the study proposes a Bayes Decision classifier associated with the finding of error probabilities and integrals in software fault prediction. This chapter discusses the fundamental software error prediction using feature and classifier data. It also discusses the proposed software error prediction with fault predictable region that includes Chernoff Bound and Bhattacharyya Bound. The proposed Bayesian decision algorithm with error probabilities and integrals of fault predictions learning model is used to predict the software faults. It works on two different bounds namely Chernoff Bound and Bhattacharyya Bound. In the Second part of the research, the study proposes an EnsembleSVM-GA learning model to predict software faults. It works on two different modules namely Ensemble-SVM and GA model for feature extraction and fault classification. The GA performs the former task and SVM the latter task. The performance of the proposed methods is tested against several other machine learning classifier over collected software fault datasets. The proposed methods and evaluated against various performance metrics: Detection Rate or recall rate, False Alarm Rate, Balance, Area Under Curve and Accuracy. The result shows that the proposed GA-Ensemble weighted SVM has higher accuracy and AUC than other methods and provides good balance than other methods. the accuracy, AUC and Balance for other methods are slightly lesser than the proposed ESVM-GA classifier for diagnosing the faults against several datasets newline