Multi-View Forests of Tree-Structured Radial Basis Function Networks Based on Dempster- Shafer Evidence Theory Mohamed Farouk Abdel Hady, Günther Palm, Friedhelm Schwenker* University of Ulm, Department of Neural Information Processing James Frank Ring, 89069 Ulm, Germany Abstract. An essential requirement to create an accurate classifier ensemble is the diversity among the individual base classifiers. In this paper, Multi-View Forests, a method to construct ensembles of tree-structured radial basis function (RBF) networks using multi-view learning is proposed. In Multi-view learning it is assumed that the patterns to be classified are described by multiple feature sets (views). Multi-view Forests have been evaluated by using a benchmark data set of handwritten digits recognition. Results show that multi-view learning can improve the performance of the ensemble by enforcing the diversity among the individual classifiers. 1 Introduction Error diversity is a fundamental requirement to build an effective classifier ensemble and therefore many definitions of classifiers diversity have been introduced e.g. ten different measures have been proposed by Kuncheva [1]. Multi-view learning is a machine learning approach where each pattern is represented by many feature sets obtained through different physical sources and sensors or derived by different feature extraction procedures leading to different types of discriminating information about the pattern. For example, a web page can be represented by different views, e.g. a distribution of words used in the web page, hyperlinks that point to this page, and any other statistical information, such as size, number of accesses, etc. The paper is organized as follows: In Section 2 the Multi-view Forests method, a new multi-view ensemble method, is explained. Results of its application to handwritten digits recognition are presented in Section 3 and finally we conclude the paper in Section 4. 2 Multi-View Forests In the multi-view learning, the input space is given by a product space X =X 1 x…x X F and data points are given by x= (x 1 , …, x F ) where x i denotes the i th feature vector. A Multi-View Forest is an ensemble of tree-structured classifiers {TC k }, k=1,…,M that are trained on the predefined subspaces {S k } (see Figure 1). The input patterns x are projected onto {S k }. The tree classifier outputs {y k } are then combined to produce the final decision y Final using a combination function such as minimum and product. _________________ * This work has been supported by the German Science Foundation DFG under grant SCHW623/4-3 and a scholarship of the German Academic Exchange Service DAAD. 307 ESANN'2008 proceedings, European Symposium on Artificial Neural Networks - Advances in Computational Intelligence and Learning Bruges (Belgium), 23-25 April 2008, d-side publi., ISBN 2-930307-08-0.