Comparing diagnostic models for CPA tumors

In this work we are concerned with comparing learning classification procedures generated from known medical cases. More precisely, we have obtained diagnosis models that discriminate between Cerebellum Pontine Angle (CPA) tumors and Otorhinolaryngological (ENT) disorders and compared them according to: accuracy, false negative minimization, and the interpretability of the model. These models are generated by means of five different learning methods: Support Vector Machine, Naive Bayes, C4.5, CART, and OneR. In order to differentiate CPA tumors from other disorders one must Perform a clinical-neuro logical examination together with expensive radiological imagery (CT and MRI) and/or invasive exams. This comparison shows how to obtain useful diagnostic models based only on clinical data, without the need of such expensive complementary exams.