Abstract:

To get the minimum misclassification of cost-sensitive classification, the algorithm of cost-sensitive extreme learning machine (CS-ELM) was proposed by using probability estimation and misclassification cost to reconstruct the classification results. Then the “rejection cost” was put into the above algorithm to further reduce the average misclassification cost. This algorithm was applied on the gene expression datasets and compared with extreme learning machine, cost-sensitive decision tree, cost-sensitive BP neural networks and cost-sensitive support vector machine. The experiments demonstrated that the CS-ELM embedded rejection cost could reduce the average misclassification cost better and could make the classification result more reliable.

Key words: extreme learning machine, cost-sensitive, majority voting, misclassification cost, gene expression data, rejection cost