Prediction of software reliability by support vector regression with genetic algorithms

This paper deals with the application of a novel neural network technique, Support Vector Regression (SVR), in software reliability forecasting. The objective of this paper is to examine the feasibility of SVR in software reliability forecasting by comparing it with various Neural Networks (NN) model and the traditional Non-Homogeneous Poisson Process (NHPP) models. In order to construct an effective SVR model, we have to setup SVR's parameters carefully. This paper proposes a new approach called GA-SVR that searching for SVR's optimal parameters by using real value genetic algorithms, and uses the optimal parameters to construct SVR models. A real failure data of a complex military computer system is used as the data set. Experimental result shows that GA-SVR outperforms the NN models and the traditional NHPP models based on the criteria of Mean Absolute Deviation (MAD), and Directional Change Accuracy (DCA).