Construct Offline and Online Membership Functions Based on SVM for Classification
Xiangsheng Rong, Ping Ling · 2007
The classification algorithm presented in this paper consists of Offline and Online Membership Functions, named as OOMF.They cooperated with each other to provide qualified class label of confidence.The offline membership function is derived from decision functions yielded by a weighted SVMs approach (WSVM).The online membership function works in the scenario where offline membership function is of low discrimination.And it is designed by a new kNN (NkNN) that is encoded with a class-wise metric.Some strategies bring computational ease: hyper parameters concerned are tuned context-dependently; training dataset is reduced by a tuning support vector clustering (TSVC); and working set of NkNN is pre-specified.We describe experimental evidence of classification performance improved by our schema over state of the arts on real datasets.