Semi-supervised Multi-kernel Extreme Learning Machine

Abstract

Extreme learning machine (ELM) is a single hidden layer feed forward neural network (SLFN). It expanded to semi-supervised ELM (SSELM) to deal with unlabeled data problem. In such a problem, labeled data is either rare or not cheap. Although SSELM has a good generalization performance, it might be influenced by heterogeneous data from different sources. It is discernible that unbalanced data issue inflicts obstacles in real-world applications including medical diagnostics and credit card fraud detection. To deal with this issue in this paper, we introduce a multi-kernel semi-supervised ELM (MKSSELM). It is more flexible to deal with discrete data from various sources. It matches diverse information from disparate sources and it shows distinction among the data. Instead of using one kernel, we optimize both ELM structural parameters and kernel combination weights. The optimization process accomplished by commanding an L1-norm as a regulation term. Meanwhile, a non-negative constraint on the kernel combination weights is used. The validity of MKSSELM algorithm is confirmed through classification results on real-world benchmark datasets. The proposed algorithm achieved better or comparable results with respect to previous approaches.