Hybrid evolutionary built decision trees for prediction of perspective cross-country skiers

In this paper we present an application of intelligent systems using hybrid evolutionary built decision trees to extract new knowledge from a small 'Cross-country skiers' database. Evolutionary built decision trees approach is a combination of one well known machine learning approach - decision trees and recently very popular evolutionary algorithms. By combining both approaches we were aiming for a method, that would combine the advantages of both approaches and should find optimal solution sooner than classical decision tree method. Such method is very suitable for application in fields like medicine and health care, where we have to combine high accuracy of the classifier and transparent knowledge representation. In the particular case of 'Cross-country skiers' database, our main goal was to develop a classifier, that would have as high prediction accuracy as possible for all three outcomes. On the other hand we were also interested in attributes that are the most important for the prediction of cross-country skier's competition potentials. The results of testing are relatively good. Overall accuracy of the classifier, which was tested on the test set ranged from 62.5% to 87.5%. In some cases we were also able to achieve relatively high average class accuracies, which was not a trivial task on database with three possible decisions and small unbalanced training set.