Study On Two Online Classification Algorithms

The two popular models of the least square support vector machine and the extreme learning machine have been paid wide range of concerns and used in the fields of the machine learning and data mining.As the development of the information techniques in nowadays society,it is general for people to deal with the data streaming.Online learning is the main way for data stream mining,and the algorithms based on the framework of the batch online learning can more effectively cope with the task of the big data.For the least square support vector machine algorithm,the proposed incremental online LS-SVMs based on chunk-by-chunk updating algorithm is an expanded version of the existing online learning LS-SVMs,which can break the limits that the classical algorithms incorporate support vectors(SVs)one-by-one.It solve the usage of the applications where data arrives in a fashion of mini-batch,by employing the block Gaussian elimination method to dynamically update the support vector set and the LS-SVMs model.To effectively and efficiently deal with the learning from imbalanced data stream that class imbalance and concept drift occur simultaneously,the novel method is proposed,known as the budget online weighted extreme learning machine model,which combined with the online sequential extreme learning machine and the weighted extreme learning machine.The budget thought is added to the model by the maximum judgment criterion to avoid the increasing memory of the information storage and structure,and the employment of matrix correction technology and Sherman-MorrisonWoodbury formula can help update the ELM model efficiently.Additionally,the validity and potential of these two models have been verified by the numerical experiments on different datasets,which can also provide a basis for further research and promotion.