Comparison of Three Neural Network Classifiers for Aphasic and Non-Aphasic Naming Data

This paper reports a comparison of three neural network models (Multi-Layer Perceptrons, Probabilistic Neural Networks, Self-Organizing Maps) for classifying naming data of aphasic and non-aphasic speakers. The neural network classifiers were tested with the artificial naming data generated from confrontation naming data of 23 aphasic patients and one averaged control subjet. The results show that one node MLP neural network performed best in the classification task, while the two other classifiers performed typically 1 2 % worse than the MLP classifier. Although the differences between the different classifier types were small, these results suggests that a simple one node MLP classifier should be preferred over more complex neural network classifiers when classifying naming data of aphasic and non-aphasic speakers.