A SOFTWARE RELIABILITY PREDICTION BASED ON RBM ALGORITHM IMPROVEMENT METHOD

Although the classic DBN has a good capability of feature extraction layer by layer and plays an important role in the area of software reliability prediction, training data still costs more time, and test error rate is relatively large. Therefore, the algorithm and network structure of DBN still has room for improvement. In this paper, the algorithms and architectures of DBN model is improved by determining the input dimension combined with self-organizing algorithm, automatically adjusting the number of hidden units and adding the support vector machine (SVM) classifier with the model output. In this way, it will not only substantially reduce the amount of information loss in each layer and the time of training data, but also improving the prediction accuracy of software reliability prediction. In order to verify the superiority of the improved software reliability prediction model, we have compared the traditional DBN software reliability prediction model with the improved one by experiment in this paper.