论文部分内容阅读
多分类器系统因其能够显著提升分类精度而引发了广泛关注。多分类器系统中各子分类器间的差异性是提升融合分类精度的先决条件。提出了一种基于证据距离的分类器系统差异性度量,同时基于该度量提出一种多分类器系统构造方法。综合了既有差异性度量、所提新差异性度量以及在训练样本集上的分类性能等多个指标,实现了多分类器系统的有效构造。实验结果表明,所提差异性度量及多分类器系统构造方法是合理的,能有效提升融合分类精度。
Multi-classifier systems have drawn much attention due to their ability to significantly improve classification accuracy. The difference between sub-classifiers in a multi-classifier system is a prerequisite for improving the accuracy of fusion classification. This paper proposes a new measure of the diversity of classifiers based on the evidence distance. At the same time, a multi-classifier construction method is proposed based on this metric. Combining the existing difference measure, the new difference measure and the classification performance on the training sample set, the multi-classifier system is effectively constructed. The experimental results show that the proposed measure of diversity and multi-classifier system construction are reasonable and can effectively improve the accuracy of fusion classification.