Assessment of Software Reliability by Object Oriented Metrics using Machine Learning Techniques

Abstract

Reliability of software is a distinct feature of reliability engineering. Software reliability models are used to assess reliability by fault prediction. Reliability is a real-world phenomenon with many related real-time problems. For easy, reliable and effective solution, many computational techniques have been developed. Though non-functional but significant feature of any Software is its “Quality”, which is not contented by many software products. Software defect prediction models with object-orientated metrics are used to measure the quality by identifying its defective classes. This research empirically analyses the defective classes by modelling with software failure data using the machine learning techniques and object-oriented metrics. Various models using Machine Learning algorithms like Logistic regression (LR), Decision Tree (DT), Support Vector Machine (SVM), Naïve Bayes Classifier (NB), k-Nearest Neighbour (KNN), Random Forest (RF) and Stochastic Gradient Boosting (GBM) have been developed. The fault proneness is detected for the classes of Marian Jureczko (MJ) Data sets. After analysis of Marian Jureczko Data set, it is found that RF provides optimum values for accuracy and ROC-AUC. The built RF model lies in outstanding category