Self-adaptive Weighted Extreme Learning Machine for Imbalanced Classification Problems.

A self-adaptive weighted extreme learning machine (SawELM) is proposed in this paper to deal with the imbalanced binary-class classification problems. SawELM calculates the outputlayer weights based on a newly-designed self-adaptive mechanism which includes the following two modules: one is to gradually reduce the weights of wrongly-classified training instances and the other is to dynamically update the outputs of these wrongly-classified instances. On 50 imbalanced binary-class data sets selected from KEEL repository, we compare the accuracy, G-mean, and F-measure of SawELM with unweighted ELM (UnWELM) and weighted ELM (WELM). The experimental results show that the newly-designed self-adaptive mechanism is effective and SawELM obviously improves the imbalanced classification performance of WELM. SawWLM obtains the significantly higher Gmean and F-measure than UnWELM and WELM. Meanwhile, the accuracy of SawELM is better than WELM and comparable to UnWELM.