Research And Application Of Data Mining Algorithm

With the rapid development of the global IT and the rapid spread of the Internet, moderninformation system enters an era of big data. Faced with excessive data and information explosionin the daily work and life, people urgent need an effective means to help them mining really neededand valuable knowledge from the massive data. Therefore, data mining algorithms has become a hotresearch topic.For the purpose to improve the time efficiency and applicability of data mining algorithms, theclassical data mining algorithms and their applications are researched in this thesis. Starting withstudying the research background and significance of research topics, this thesis introduces relatedconcepts, meanings and basic techniques of data mining; mainly studies the density-basedclustering algorithm DBSCAN, rough set attribute reduction algorithm and BP neural networkalgorithm.For clustering, in order to reduce the algorithm execution time of DBSCAN algorithm, agrid-based fast DBSCAN algorithm named GF-DBSCAN is proposed through improving DBSCANalgorithm, and its good performance is verified by simulation experiment.In this thesis, a short-term power load forecasting strategy based on a variety of datamining technology is also proposed; the strategy includes a hierarchical prediction model and themethods used by each layer. The first layer uses DBSCAN algorithm to cluster data, identify theoutliers and then modify them effectively. The second layer effectively selects the minimum set ofattributes which associate with decision-making by using attribute reduction algorithm based onrough sets to guarantee the efficiency of mining. The third layer uses BP neural network algorithmto get the predictive value by using its complex multi-layer, multi-node network structure and errorsback-propagation fixed ability. The results of doing computation and analysis on actual data havedemonstrated the superiority of the proposed strategy in improving the accuracy of short-termpower load forecasting.In addition, this thesis proposes a personalized push service technology scheme based onGF-DBSCAN algorithm, which can be applied to e-commerce operation platform. Firstly, it usesGF-DBSCAN algorithm to mine users’ basic information for classifying the users into certainclusters. Secondly, it uses top-k sorting algorithm to analyze users’ interest of each cluster. Finally,it pushes the analysis results to the corresponding users. The applied effect of an online bookstore has shown the practicality of GF-DBSCAN algorithm.The thesis has done beneficial research work on data mining algorithms and their applications.