Classification Tress and Neural Networks are two popular approaches to Pattern Recognition problems. Both these approaches are combined in NeurlTree which uses a Multi-layer Perceptron (MLP) at each decision node of binary classificatio tree to extract non-linear features. NeuralTree exploits the power of tree classification using appropriate local features obtained by trained Neural Networks at internal nodes. This approach has been successfully applied to recognize hand-written isolated digits. The proposed method acheives significant decrease in error-rate compared to other classical methods and the size of NeuralTree classifier is also small compared to that of Classification and Regression Tree (CART).
展开▼