Abstract
Usually, algorithms for machine learning during the classification return a single class for a given object. Many of the systems do not estimate a reliability of their answer. In the article a method is presented that returns multiple classes as possible. The method also gives the user an estimation of the answer's reliability. Additionally, the method enables also classification in domains where one example can belong to more than one class. The described ideas are tested on a real medical domain - rheumatology. The results are compared with the results of the classical algorithms for machine learning and with the results of general practitioners.