Data stream mining

Data streaming is one area of data mining that has been studied extensively. One problem of data streaming is to detect noise and random shapes when clustering, where basic K-Means usually fail. Some researchers suggested density based clustering according to a decay function; one typical example is D-Stream. However, its universal decay factor and cluster on a fixed interval do not achieve optimal efficiency regarding to space and time complexity. In this report, we made an attempt to improve both space and time complexity of D-Stream. Our integrated work DCC-Stream follows conventional online-offline approach in stream mining. We describe our algorithm as two parts: online and offline parts. Online part accumulates historical data as synopsis information and makes use of two sentinels to detect whether offline parts should be invoked. Offline part contains two separate parts, one is responsible for updating density, the other is for clustering. The experimental evaluation shows that our algorithm achieves both significant improvements on time and space complexity. The results show time usage is greatly reduced while maintain similar purity. In addition, the algorithm also achieves better space usage.Bachelor of Engineering (Computer Engineering