A New Model of Combining Multiple Classifiers Based on Neural Network

Classification is a very important part in the domain of pattern recognition. However, single classifier has many defects, such as very finite applicability and low accuracy. Combining multiple classifiers can overcome the defects. Method of combining the classification powers of several classifiers is regarded as a general problem in various application areas of pattern recognition, and a systematic investigation has been made. Possible solutions to the problem can be divided into several categories according to the levels of information available from the various classifiers. The existent methods for the combination of multiple classifiers may be classified into three frame works, i.e. linear opinion pools, winner-take-all and evidential reasoning. In this paper, the authors develop a new model of combining multiple classifiers based on neural network. Do not merely use the neural network as a simple classifier, but to take full advantage of its noise high affordability advantages, we use neural network as a model for training framework for the fusion rules. Experiments on the UCI data sets show that it can improve not only the accuracy of classification but also its applicability.