Research On Feature Representation And Attribute Selection Algorithm For Multivariate Time Series

Time series refers to a set of observation time set by time,and widely used in financial,medical and other fields,while the time series contain high dimension,large scale,complex structure and noise interference and other shortcomings,in order to reduce the data processing,This paper proposes a new method for improving the accuracy of time series data mining.Based on the related literature,this paper puts forward some improvement methods for MTS in the feature representation and feature selection(attribute selection).Work as follows:(1)Corrected the shortcomings of the MCPCA multiple times the time series feature representation algorithm.Before introducing the improved algorithm,basic idea of the CPCA algorithm and the MCPCA algorithm are introduced.And then points out the problems existing in the MCPCA,that is,all sub-modules have the same weight,propose an MTS feature representation algorithm based on class separability weighting.In order to highlight the weight of different sub-modules and improve the classification accuracy,class separability is used in the sub-module weighting of MTS.By assigning weights to the different sub-modules by maximizing inter-class discretization and intra-class divisiveness ratios,Weighted sub-modules can highlight the local information,improve the contribution of local information on the classification.(2)Aiming at the problem that the accuracy of the existing MTS attribute selection algorithm is not high and the time complexity is high,a MTS attribute selection method ACDR based on correlation density is proposed,and the fast clustering algorithm is applied for the first time in the MTS attribute selection,attribute selection process is considered as clustering process,various central sample points are chosen as representative attributes.At the same time,in order to solve the problem of MTS unequal length and high computational complexity,the correlation measure method is introduced,and the correlation matrix is used instead of the original MTS matrix.Finally,in the attribute reference index chart drawn according to the ACDR algorithm,the inflection point in the graph is defined as the point of the second order change rate of the attribute reference index,and the attribute before the inflection point is selected as the characteristic attribute,and the attribute subset is formed.This algorithm is an effective MTS attribute selection algorithm.