Combining Methods to Stabilize and Increase Performance of Neural Network-Based Classifiers

In this paper we present a set of experiments in order to recognize materials in multispectral images, which were obtained with a tomograph scanner. These images were classified by a neural network based classifier (Multilayer Perceptron) and classifier combining techniques (Bagging, Decision Templates and Dempster-Shafer) were investigated. We also present a performance comparison between the individual classifiers and the combiners. The results were evaluated by the estimated error (obtained using the Hold-Out technique) and the Kappa coefficient, and they showed performance stabilization.