A Hybrid Technique for Software Reliability Prediction

Abstract

Reliability is an important factor of software quality. The accurate prediction of software reliability is a challenging task. There exist many reliability models to predict the reliability based on software testing activities. There are many software reliability growth models (SRGMs) developed to predict the reliability but they have many unrealistic assumptions and they are also environment dependent. The accuracy of the models is also questionable. In this paper we have used a time series approach for software reliability prediction. We have used an ensemble technique called hybrid ARIMA (ARIMA + NN) for prediction of software reliability based on real life data on software failures. This paper also gives a comparative analysis of forecasting performance of hybrid ARIMA, and ARIMA models. Empirical results indicate that a hybrid ARIMA model can improve the prediction accuracy.