Optimized selection of weak methods for the classification of endoscopic images using an ensemble classifier

In the past we developed an ensemble classifier in order to improve the accuracy in terms of the classification of endoscopic images. However, since we have a variety of feature extraction methods for the construction of a weak method set at hand, the number of different possible weak method combinations for the ensemble is quite huge. In order to address this issue we propose two different methods which aim at determining a set of weak methods which delivers an optimal overall classification rate. While the first algorithm determines optimal weak method candidates by a rating based on the candidate set, the second algorithm aims at constructing a set of statistically significant methods in order to increase the diversity of the ensemble. Based on previously developed methods, we evaluate the proposed methods by comparing the overall rates achieved by the respective combinations to the overall rate achieved by the best possible combination. We show that the proposed selection algorithms are able to find the best combination of methods or at a least competitive one.