Research On Incremental Classification Algorithm Of Multiple Center Vector Base On Minimun-distance

Classification is an important part of data mining,especially with the advent of big data era,one of the focused subjects of data mining is how to improve the traditional classification algorithm, so that it can classify the big data.Based on the existing algorithms, a new incremental classification algorithm of multiple central vector base on minimun-distance is presented in this dissertation, which puts forward the concepts of regional division and multiple centrals.The algorithm can be divided into four phases. In the first phase, according to K-means algorithm to cluster the training data to form diverse subsets, then eliminate the overlap of class space by adjusting the space between classes. In the second phase, the data space is divided into stability region, overlap boundary region and unknown region. In the third phase, the samples that fall into the stability region and the overlap boundary region are processed in "Direct Classification" and "Most of the Delegates Classification" respectively. Next mutilple centre vector is proposed to classfy the incremental data which fall into the unknown region. The fourth phase is process of selecting sample. In order to save center vector so that available above presented classification method, the samples of near center region are selected as examples, the other are selected interval.The two experiments on combines data which is normal distribution and real data get from UCI website indicate that compare above presented algorithm with the algorithm from literature[38],the classification accuracy is nearly same, but the demand of space and the time is decreased in some degree, it is significant for big data classification.