Class-wise multi-classifier combination based on Dempster-Shafer theory

Multi-classifier combination based on Dempster-Shafer theory of evidence has demonstrated it´s superior performance. In the approach based on Dempster-Shafer theory, the basic probability assignments for evidence is usually derived from classifiers´ global performance. However, our study discovered that while using classifiers´ global performance as basic probability assignments doesn´t necessarily improve performance under some circumstances, the alternative approach using classifiers´ class-wise performance as basic probability assignments does improve the classification performance and outperforms the traditional one based on classifiers´ global performances. Basic probability assignments based on classifiers´ class-wise performances results in more accurate calculation of beliefs, thus boosts the combinator´s performance.