Software fault prediction performance in software engineering

Software fault prediction improves software quality and testing efficiency by early identification of faults. Classification models using code attributes are constructed and used for prediction. This paper is a study of software fault prediction using Multi-Layered Perceptron, Bayesian Network and Naive Bayes classifier and their comparison by showing predictive and comprehensible performance. A framework is proposed for software fault prediction and applied on 10 public domain data sets from NASA PROMISE Repository. The predictive accuracy is observed, which supports the view that software metric based classification is useful. Furthermore, the accuracy is increased up to 85% or more by means of selecting methods and code attributes of data sets. The results are compared in terms of True Positive Rate (TPR) and False Positive Rate (FPR). Output shows that the neural network classification models are more superior to the other network models.