Incremental Support Vector Clustering Algorithm

摘要

Support Vector Clustering(SVC) algorithm is a extended method to unsupervised learning problem from Support Vector machine that has mainly applied in supervised learning problem such as classification and regression. According to kernel mapping, data points are mapped from data space to a high dimensional feature space. Then after the sphere analysis in feature space, SVC finds Support Vectors(SVs) that describe the cluster boundaries. Therefore the SVC is a kernel and boundary based cluster analysis. Main advantages of the SVC method are detection of the arbitrary shaped cluster boundary and robustness about the noise without preknowledge about the data distribution. On the other hand, due to the high computation complexity, it needs high cost about learning. As the result if there is the need for the analysis about the general databases and applications which data updates are collected and applied to the database periodically, we choose either disregard new input pattern and keep the previous learned results or disregard previous high cost learned results and learn newly after adding the new data into previous database. In this paper, We maintain the merits of SVC and extend it to the incremental learning for periodically updated database and variety datamining application. The new incoming data process SVC block-by-block learning, and then processed results are combined into the previous SVC results, so that it shows the overall results of data cluster. We demonstrate that the proposed incremental clustering algorithm produces high-quality clusters and identify meaningful patterns with new input data stream.