The Local Outlier Mining Algorithm Based-on Conditional Cumulative Holoentropy And Global Neighbourhood

Outlier mining is one of the important branches of data mining.Its main task is find objects that deviate from other data characteristics.In our daily life,most things are common.But we not ignore the existence of many unusual objects.These objects make a significant sense.At present,various methods have been proposed for outlier mining by domestic and international researchers.In this paper,The subspace clustering and local outlier mining algorithms are studied for the problem of poor performance of local outlier mining algorithm.The main contents of this paper include the four aspects.Firstly,the research background and significance of local outlier mining are analyzed,and a outlier mining survey of domestic and international research advances is also made.The classical subspace clustering and outlier mining algorithms are analyzed and introduced in detail.Secondly,Aiming at the problem that the subspace clustering effect of CMI algorithm is poor.it is used the Conditional Cumulative Holoentropy to estimate the subspace clustering,a local outlier mining algorithm based on Conditional Cumulative Holoentropy is proposed.so as to select the optimal clustering subspace.Then,the LOF algorithm is used to perform outlier detection in the optimal clustering subspace.Thirdly,combined with the HICS algorithm,and aiming at the problem that the Lo OP algorithm is not accurate on the global outlier detection,the use of global neighborhood instead of local neighborhood for outlier detection,a global neighborhood based on the local outlier detection algorithm is proposed.At the same time,the proposed local partial outlier probability parameters are introduced and analyzed,and its effectiveness is analyzed for verify the algorithm has achieved better detection results.Finally,the two algorithms are implemented in this paper on the UCI real data set and the virtual data set,compared with the comparative experiment respectively.The accuracy and efficiency of the two algorithms are verified by the experiment.