Algorithms And Application Research On Multi-indices Association Analysis And Prediction

This paper is done to meet the requirements of information analysis and prediction for the China National Power Corporation Information Management System. The main work and achivements are follows.Firstly, in order to make an analysis and prediction of multi indices correlation, the continuous data must be discretized . After serveral methods of discretizing contionuous data are analyzed in this paper, multi-indice data of GuoDian information system are discretized according to the methods of fields discretization based on density distribution clustering. As data mining is difficult to be done in a single machine with the increment of bussiness data, an algorithm of mining FP tree in parallel, FPTDP, is covered in this paper.Secondly, as new data are added to the original dataset continually, one must take consideration in making use of results which were acquired in the process of mining the original dataset. So incremental mining is one of the topics in this paper. Because the traditional incremental minings , at their worst, need to rescan the original dataset DB, which decreases their the performance. An algorithm of incremental mining based on dictionary tree,DTARIDMA, is discussed after the analysis of serveral traditional incremental mining. Its parallel mining algorithm is also given.Thirdly, it is proved that the FPTDP algorithm is valid and can lower the performance requirement of a single machine by means of making experiments in the paper. With experiments we compare the performance of the DTARIDMA algorithm and traditional incremental mining. Association rules mining is also made on multi-indice of GuoDian information system. It is proved that the DTARIDMA algorithm is a valid and applicable multi-indice association rules mining algorithm.