Multiple kernel learning-based multi-functional nearest-neighbor classification

One of the most important issues for the development of classification tasks is how to design a powerful classifier with strong generalization capability. Multi-functional nearest-neighbor (MFNN) has recently proven to generalize the different versions of nearest-neighbor classifier by choosing appropriate similarity measures and aggregators, and has been demonstrated to enhance the more comprehensive adaptivity and higher accuracy. In this paper, the multiple kernel learning-based multi-functional nearest-neighbor classification approach (MMNN) is proposed to improve the performance of MFNN. In terms of the superiority that kernel-based fuzzy rough set have, kernel functions can be used to compute fuzzy similarity relations. MMNN is a reformative framework to combine a particular multiple kernel learning method with a flexible nearest-neighbor classifier named MFNN. Meanwhile, the results of experiment show that the proposed method outperform other state of the art algorithms.