The Research Of Data Stream Mining Algorithms On Smart Mobile Devices

For the problems of more effective to get useful information from data stream, manyscholars have carried out extensive and in-depth study. With the development of smartmobile devices and wireless network technology, people are more inclined to get miningresults in anywhere or anytime, namely UDM. However, traditional data stream miningalgorithms cannot be used in smart mobile devices, because of the emergence of newissues, like resource constraints. Therefore, for the purposes of applying data streammining algorithms in smart devices to meet the need of UDM, we have done the followingaspects of research in this paper:First, in order to better use of data stream mining algorithms on smart mobile devices,a data stream mining algorithms framework MMF is developed in this paper. According tothe different environment or needs of users，the users can select with stream mining serveror smart device to accomplish the task of stream mining to meet the needs of the differentmining quality. In stream mining server, traditional data stream mining algorithms can beused; In smart device, at the same time to ensure the mining quality, a adaptive componentis used to adjust algorithm or device itself to solve the issues such as resource-constrained.Secondly, in order to improve the clustering quality of the data stream clusteringalgorithms in the smart mobile devices, as well as the adaptability of resources, anarbitrary shape of data stream clustering algorithm based on MMF have been proposed.Sliding window model is used, new micro-cluster feature structures are maintained inmemory, time decay model is used to deal with the problems of historical data and datastream evolution; a new micro-cluster cycle detection mechanism are proposed to improvethe efficiency of algorithm; adapting strategies to adapt to the limited equipment resources;the classic DBSCAN algorithm is used to discover clusters of arbitrary shape.Finally, in order to improve the classification accuracy and the adaptability of theresources of the data stream classification algorithm in the smart mobile devices, a datastream classification algorithm based on MMF have been proposed. Semi-supervisedk-means algorithm based on rough set is used to train new classifier models and the methods of k-nn and vote are used to classify unlabeled data to solve the problems ofpartially labeled data and classification uncertainties; a new detection mechanism isproposed to solve the concept-evolution problem; ensemble classifier model is used toaddress the concept drift problem; a series of adaptive strategies are used to solve theproblem of limited resources.Verified through experiments on real and artificial data, the two algorithms areefficient, reliability and adaptability of resources.