Data Stream Classification Based On RVFL

With the continuous development of communication technology,the data connection between various terminals becomes more and more frequent.There are a lot of data stream between terminals,people not only need to obtain the surface data embodied in the data stream,but also want to mine the deep data hidden in the data stream.However,due to the real-time characteristics of data stream,it is difficult to obtain it again,the traditional classification method of machine learning training based on static data set is not suitable for high-speed and large amount of data stream.Therefore,for the classification process of data stream,we need a method to process the data stream quickly.And there are high-dimensional,linear and nonlinear data in the data stream.The sample concept changes as the data stream changes over time,which will lead to poor classification results.This phenomenon is called concept drift.How to classify such data streams quickly and accurately is an important issue.Based on the above background,because there are many high-dimensional data in the data stream,in order to avoid dimension explosion,a combination of global embedding using subspace-based angle optimized global embedding(AOGE)and principal component analysis(PCA)is used in this thesis.Choose the minimum spacing in the class by using AOGE cosine projection method and choose the maximum inter-class distance by PCA method,which optimized the concept drift detection dimensionality reduction process and judged the concept drift.By comparing the results of experiment,the data after dimensionality reduction by angle liner discriminant embedded(ALDE)are selected to be fed into the random vector functional link(RVFL)classifier for training.At the same time,the concept drift detection framework of data stream with AOGE and PCA is used for concept drift detection.The dimensionality reduction process and concept drift decision are combined so that the data can be carried out simultaneously in the processing flow.When the data stream does not produce concept drift,the algorithm uses the original training model.Once the concept drift occurs,the new data is added to the training for model iteration,which improves the efficiency of the algorithm.In this article,we discuss RVFL,Extreme Learning Machine(ELM),and Support Vector Machine(SVM)classifiers,Finally,a method of data stream classification based on RVFL classifier is proposed.