Predicting Time between Software Failures Using ISGNN
Aiguo Li, Dashan Qiu, Zhanhuai Li · 2006
Neural networks methods have been used in prediction of time between software failures. However, most of the existing methods have some shortages, including requiring much complex computing and requiring advanced users to set their network structures and many parameters. Iteration learning selfgenerating neural networks (ISGNN) is an improved selfgenerating neural networks (SGNN). It has inherited the advantages of SGNN, for example, users do not need to set network structures and parameters and it has better classification precision. We propose a scheme based on ISGNN to predict time between software failures in this paper. Two real failure datasets are used in experiments in this paper. Experimental results show that ISGNN is easy to use and its training time consume is about one sixth of BP (Back-Propagation) networks’, and the MAE (mean absolute error) and RMSE (root mean square error) of