Mining Inter-transaction Association Rules From Multiple Time-series Data

With the rapid economic development,there will generate numerous time series data in many different areas,such as finance,medicine,geology,meteorology,e-commerce,sensor networks and other fields.It is a significant orientation of research for mining time series to predict the hidden relation in time series,which is very important for the production and life.Because of the mass,realtime,continuity of time series data stream,it is difficult to use the traditional association rule mining method to mine the time series directly.At present,there are some research results about the association rules mining of time series data stream.These studies have been studied on the single time series and the mining of association rules in transaction.However,the research on the inter transaction association rules mining of multiple time series is less.And the timeliness,sequential is ignored and incremental mining association rules can not be achieved in most of the research on association rules.In this paper,a algorithm of piecewise linear representation which is named ITEO is used to compress the time series data.In the time series clustering algorithm,the IK-Means algorithm is proposed to select the initial clustering point.This algorithm not only increases the class spacing between clusters but also reduces the uncertainty of random selection of initial cluster centers.This topic uses 3 parameters(the length of time,numerical intercept divided by minimum point value,slope segment)to represent mode,normalized the similarity measure.In the mining process of association rules,this paper design the Inter-Transaction Association Rules from Multiple Time-Series List(IAMTL).Based on the storage structure,the association of multiple time series in a fixed time T can be represented,adding fixed time T constraints in time series association rules.It enhances the order of association rules.I-IAMTL algorithm adapts modified support and confidence to mining association rules incremental.The accuracy of the prediction of association rules is verified by the method of prereqiusite and the consequent windows(PCW).On the actual industrial data and stock data,by comparing the existing algorithm with the IAMTL and I-IAMTL algorithm,it indicates the validity of our proposed algorithm.