New feature selection methods for qualification of the patients for cardiac pacemaker implantation

Implantation of a cardiac pacemaker is a complicated procedure. The success of the procedure depends directly on the proper classification of patients and the choice of the type of pacing. Machine learning algorithms can support this process. The most important element of these is the feature selection process. In this paper we present the results of our own implementation of feature selection methods, working on the electrocardiological datasets of 4316 patients with severe heart rhythm disorders and qualified for pacemaker implantation. For the research, we chose the two most promising algorithms (CFS and Chi-square). In all cases it was possible to reduce the initial set of attributes by 60%. Due to the reduction of the search space the number of generated decision rules was decreased by factor of 6-10. Because of this, practical cardiological validation of rules is easier and faster, more general rules adapt better for recognition of new cases and computational effort is reduced, which was confirmed in clinical practice.