Time density oriented clustering mechanism

摘要

Clustering is an important task in data mining area, especially in the area of continuous stream of data, i.e. ?data stream?. However, some characteristic of this kind of data is neglected during the existing clustering approaches. The similarity in temporal dimension between entities is underestimated. Forgetting mechanism is adopted to remove the old patterns to save computation resources. However, these old history may be used in close future. For these issues, a time density oriented clustering mechanism is proposed. The similarity in temporal dimension of data is considered for clustering. Only the entities in close time range could be grouped together. In this clustering mechanism, the concept of time density is introduced to measure the similarity in temporal dimension. Analysis and simulations show that the proposed clustering mechanism is effective and efficiency for data stream.