The Multiple-classifier Fusion Model Based On Niche Genetic Algorithm

With the in-depth study of the classification, the type and design methods of the classification are increasing, But in practice, it is difficult to choose the best classifier and the best design method. Because, all kinds of classifiers have their advantages and scopes of application. Therefore, in practice, people select classifiers almost based on the experience and the actual conditions.As we all know, "Two heads are better than one". The classification problem can learn from this approach. The advantages of multiple classifiers can be combined to comprehensively analyze the same classification problem, and achieve a scientific decision in finally. Most of the existing experiments indicate that the performance of multi-classifiers fusion is superior to single classifier.Fuzzy Integral is an aggregation tool in multi-classifier fusion, which can improve the accuracy of classification and the robustness of systems. In multi-classifier fusion based on fuzzy integrals, the performance of fusion systems is decided by the selection of fuzzy measures. If the fuzzy measures are well defined, the performance of fusion systems can be improved distinctly. However, if the fuzzy measures are badly defined, it is possible that some classifier's performance is higher than the fusion systems'. Therefore, people want to improve the performance of fusion systems, and seek the better fuzzy measure in essentially.In this paper, firstly, the common classifiers and traditional fusion methods of multi-classifiers fusion are introduced and analyzed. From the conclusion, their superiority and weakness are discovered. Secondly, the model of multi-classifiers fusion based on fuzzy integrals is structured, and the common methods to define the fuzzy measure are introduced. The principles and the main steps of the genetic algorithm are described, and the advantage and disadvantage of GA are analyzed. To improve the design elements of the genetic algorithm, the niche genetic algorithm is introduced to define the fuzzy measures. Finally, two different types of simulation experiments are conduced to confirm the feasibility and effectiveness of the niche genetic algorithm. In addition, the experimental results clearly shows that the niche genetic algorithm can improve the classification accuracy of the multi-classifier fusion system, compare with the traditional methods.