一种基于贝叶斯方法的多分类器组合优化算法

An Optimization Algorithm for Combining Multiple Classifiers Based on Bayesian Approach

提出了一种基于贝叶斯方法的多分类器组合优化算法和阈值改进方法。首先,计算分类器对各个类别的置信度。然后,以各分类器的置信度为先验概率,采用向量求和将各分类器的先验概率向量进行组合,得出最终输出向量,最后通过优化阈值提高综合分类器识别精度。在此后的实验数据表明:该算法具有方法简单、运算速度快、分类精度高等优点。

This paper proposed an optimization algorithm for combining multiple classifiers based on Bayesian aproach and the method to improve the threshold.Firstly,these classifiers's confidence was computed.Secondly,these classifiers's confidence was used as pre-probability,and these pre-probability's vectors were combinated by adding these vectors.The final output vector was obtained.Lastly,the compositive classifier's recognizing precision was improved by threshold optimization.Experimental data showed that this arithmetic has simple,fast and high classification accuracy.