IPK Gatersleben, Pattern Recognition Group, Gatersleben,.pdf

Relevance learning for mental disease classification Barbara Hammer 1 , Andreas Rechtien 2 , Marc Strickert 3 , and Thomas Villmann 4 (1) Clausthal University of Technology, Institute of Computer Science, Clausthal-Zellerfeld, Germany, hammer@in.tu-clausthal.de (2) University of Osnabrück, Department of Mathematics/Computer Science, Osnabrück, Germany, arechtie@uni-osnabrueck.de (3) IPK Gatersleben, Pattern Recognition Group, Gatersleben, Germany, stricker@ipk-gatersleben.de (4) University of Leipzig, Clinic for Psychotherapy and Psychosomatic Medicine, Leipzig, Germany, villmann@informatik.uni-leipzig.de Abstract. In medical classification tasks, it is important to gain in- formation about how decisions are made to ground and reflect therapies based on this knowledge. Neural black box mechanisms are not suitable for such tasks, whereas symbolic methods which extract explicit rules are, though their tolerance with respect to noise is often smaller since they do not rely on distributed representations. In this article, we test three hy- brid prototype-based neural models which combine neural representations with explicit information representation in comparison to classical decision trees for mental disease classification. Depending on the model, informa- tion about relevant input attributes and explicit rules can be derived. 1 Introduction An important problem in psychotherapy is to identify border-line patients which are heavily disturbed since a specific treatment and therapy management is nec- essary for these patients [2]. This classification has to take place before the treatment. Commonly, therapists make their decision depending on the clinical impression of the patients within an initial session. As an alternative which does not require human intervention, the self-report questionnaire Borderline Person- ality Inventory (BPI) has been developed which allows automatic classification of border-line patients [3]. The questionnaire is answered by the patients an