Supervised multi-class classification with adaptive and automatic parameter tuning

In this paper, a classification framework is developed to address the issue that empirical determination of the parameters and their values typically makes a classification framework less adaptive and general to different data sets and application domains. Experimental results show that our proposed framework achieves (1) better performance over other comparative supervised classification methods, (2) more robust to imbalanced data sets, and (3) smaller performance variance to different data sets.