An efficient ensemble classification method based on novel classifier selection technique

Individual classification models have recently been challenged by ensemble of classifiers, also known as multiple classifier system, which often shows better classification accuracy. In terms of merging the outputs of an ensemble of classifiers, classifier selection has not attracted as much attention as classifier fusion in the past, mainly because of its higher computational burden. In this paper, we propose a novel technique for improving classifier selection. In our method, the simple divide-and-conquer strategy is adapted in that a complex classification problem is divided into simpler binary sub-classification problems. The proposed ensemble classification technique has the following advantages: f) it requires much less computation than the existing ensemble classification methods. 2) It improves overall classification accuracy. 3) It is also suitable for tackling the classification problems which have a relatively large number of target classes. We conduct extensive experiments on a series of multi-class datasets from the UCI (University of California, Irvine) repository and compare several well-known classification approaches. The experimental results demonstrate the advanced performance of the proposed method.