Predicting and classification of software reliability using ensemble learning

Abstract

Software reliability plays a pivotal role in determining the overall system reliability and is an inescapable factor when assessing the integrity of any software product. When it comes to creating mission-critical software like software for space exploration, the health sector, scientific calculation, the aerospace industry, etc., where the need for high reliability is paramount, we encounter numerous challenges that need to be effectively addressed. Accurate prediction of software reliability ensures software quality, which ultimately builds the confidence of the customer in the software they are using.  Machine learning, particularly the ensemble method, is very important to solve these prediction problems. This research develops an ensemble learning technique for software reliability prediction. Ensemble methods, which are a combination of more individual ML models are used in this research. Bagging, Boosting, and stacking techniques are applied for classification and prediction. Prediction is used to predict the failure time of the software based on the Mean Time Between Failures (MTBF). Musa, J.D’s benchmark dataset on Software reliability is used for prediction. The classification is used for classifying the software for the presence of defects or not. NASA dataset is used for classification. The proposed model achieves94% prediction and 97% classification accuracy.