Artificial Neural Network Applications for Software Reliability Prediction

Chapter 4Prediction of Time Between Successive Software Failures

These days software has wide number of applications and software failures may lead to system failure and severe consequences. Therefore, it is essential to monitor and ensure software reliability. There exist a number of software reliability models to predict software reliability in terms of number of failures and time between failures (Lyu 2006, Xie 1991). These models can be classified in two categories as statistical models (parametric models) and artificial intelligent models (non-parametric models).

The parametric models are based on certain assumptions regarding software development/testing process, software faults characteristics and nature of software projects whereas the non-parametric models do not make such assumptions. Both parametric and non-parametric models require failure history for reliability estimation and prediction. However, it is observed that non-parametric models, in general, provide better prediction accuracy than parametric models. On the other hand, parametric models provide better understanding of failure behavior through various parameters and indices. Artificial neural network (ANN) is a non-parametric model and many ANN models have been developed to predict time between failures or cumulative number of failures in software (Tian 2005b, Ma 2011, Tian 2005c, Khoshgoftaar 1996). In this chapter, ANN is used for predicting time between failures. The time between successive failures indicate ...