Dynamic confidence values selection — Experimental studies

The machine learning methods are often used in the development of the effective medical decision support systems. One of the latest trends in data mining is the ensemble selection. In this paper, we present the algorithm of the dynamic confidence values selection, which is dedicated to the binary classification task. In the experiment we use Support Vector Machine, k-Nearest-Neighbors, Neutral Network and Decision Trees models as base classifiers. Experiments on several publicly available medical diagnosis data sets verify the effectiveness of the proposed algorithm. The results demonstrate that the dynamic confidence values selection outperforms the ensemble classifier built with all base learning models.