A Combined Model for Software Reliability Prediction Based on Neural Network
Faming Gong, Kewen Li, Kang Pei Zhao · International Journal of Applied Mathematics & Statistics/International journal of applied mathematics and statistics · 2013
The single methods are usually combined by the method such as average, but when single model can?t adapt to the subtle changes, the prediction error will increase dramatically. In this paper, we propose a new combined prediction model based on neural network. First we combine prediction results of different single models linearly, including mean method, median method and weight method. Then we train the neural network by taking the linear combination results of the three methods as the input, and taking the actual value as the output. The method reduces the risk of the loss of prediction accuracy which results from the deviation of single model accuracy in the traditional combined model of neural network. Simulation experiment shows that the model can reduce the prediction risk of single neural network and improve the prediction accuracy.