数据驱动的软件可靠性模型研究

raditional software reliability growth models (SRGMs) are generally based on several impractical assumptions, which to a large extend limits their applicability and accuracy. In recent years, data-driven approach to software reliability modeling has attracted a lot of attention, and several artificial neural network (ANN) based and support vector machine (SVM) based software reliability models (SRMs) have been proposed in the literature. Data-driven SRMs require no assumptions on the properties of software faults and software failure process, thus they appear to have wider applicability compared with SRGMs. For data-driven SRMs, software failure data used have great impact on model prediction accuracy; however, to the best of our knowledge, this issue has not been studied in the literature. In this paper, an SVM-based SRM is proposed. It is also demonstrated that for data-driven SRMs accumulative software failure data rather than inter-failure data should be used, and recent failure data rather than all historical failure data should be used. A genetic algorithm (GA) based algorithm for optimizing model parameters is proposed. Based on three failure data sets published in the literature which are taken from real-life software projects, comparative studies of the proposed SRM and existing data-driven SRMs are conducted. Results show that the proposed SRM seems to have the highest prediction accuracy.