A New Real-time Clustering Algorithm

Clustering is one of most important building fields in data mining and in machine learning in general. Most clustering algorithms is designed for off-line (or batch) processing, in which the clustering process repeatedly sweeps through the set of data samples in order to capture its underlying structure in a compact and efficient way. However, with the continuous increment of set of data samples, many recent applications require that the clustering algorithm should be online, or incremental in order to save time and improve efficiency of the algorithm. In this paper, an OPTICS based incremental clustering algorithm is put forward. It adopts reachability-polt to reflect the underlying structure of data sets. Through the experiment, it shows that the algorithm not only inherits the advantages of the classical OPTICS on clustering accuracy, but also possesses the functions of online clustering. Compared with the classical OPTICS, the proposed algorithm effectively reduces the time cost of clustering.