Cost-Sensitive Extreme Learning Machine

ELM is an effective machine learning technique, which works for the "generalized" single-hidden-layer feed-forward networks. However, like original SVM, ELM and majority of its variants have been extensively used in classification applications. Compared to SVM, ELM achieve optimal solutions and require lower computational complexity. More and more researchers have been attracted by ELM due to its fast learning speed and excellent generalization performance. Traditional ELM presumes higher accuracy based on the assumption that all classes have same cost, and the sample size of each class is approximate equal. However, the assumption is not valid in some real cases such as medical diagnosis, fault diagnosis, fraud detection and intrusion detection. To deal with classification applications where the cost of errors is classdependent. we propose a cost-sensitive ELM Experimental results using classification data show that CS-ELM is effective.