Combining classifiers in software quality prediction: A neural network approach

Abstract

Software quality prediction models seek to predict quality factors such as whether a component is fault prone or not. This can be treated as a kind of pattern recognition problem. In pattern recognition, there is a growing use of multiple classifier combinations with the goal to increase recognition performance. In this paper, we propose a neural network approach to combine multiple classifiers. The combination network consists of two neural networks: a Kohonen self-organization network and a multilayer perceptron network. The multilayer perceptron network is used as Dynamic Selection Network (DSN) and Kohonen self-organization network is served as the final combiner. A case study illustrates our approach and provides the evidence that the combination network with DSN performs better than some other popular combining schemes and the DSN can efficiently improve the performance of the combination network.. © Springer-Verlag Berlin Heidelberg 2005.