Classification Algorithm Based on Weighted SVMs and Locally Tuning kNN

This paper presents a new classification algorithm consisting of a Weighted SVMs approach (WSVM) to identify data class, and a locally Tuning kNN (TkNN) to address the rejected case. Basic SVM of WSVM is equipped with weights derived from SVM output distribution to demonstrate decision confidence. These weights influence label assignment. TkNN handles the difficult cases rejected by WSVM. It works in the neighborhood that is developed by a locally informative metric. SVM decision interfaces helps to define the new metric. Hyper parameters of basic SVM are learned context dependently. We present experimental evidence of classification performance improved by our schema over state of the arts.