A generic data-driven software reliability model with model mining technique

Complex systems contain both hardware and software, and software reliability becomes more and more essential in system reliability context. In recent years, data-driven software reliability models (DDSRMs) with multiple-delayed-input single-output (MDISO) architecture have been proposed and studied. For these models, the software failure process is viewed as a time series and it is assumed that a software failure is strongly correlated with the most recent failures. In reality, this assumption may not be valid and hence the model performance would be affected. In this paper, we propose a generic DDSRM with MDISO architecture by relaxing this unrealistic assumption. The proposed model can cater for various failure correlations and existing DDSRMs are special cases of the proposed model. A hybrid genetic algorithm (GA)-based algorithm is developed which adopts the model mining technique to discover the correlation of failures and to obtain optimal model parameters. Numerical examples are presented and results reveal that the proposed model outperforms existing DDSRMs.