Efficiently Mining Association Rules from Time Series

Traditional association rules are mainly concerned about intra-transactional rules. In time series analysis, intra-transactional association rules can only reveal the correlations of multiple time series at same time. It is difficult to forecast the trend of time series. In this paper, it is studied the mining problem of inter-transactional association rules in time series with which the trend can be forecast by the time difference between the prerequisite and the consequent in a rule. A new algorithm, ITARM, for inter-transactional association rules mining is presented. It uses a compact FP-tree based and divide-and-conquer approach. After the frequent 1-itemsets is produced, the algorithm separately uses them as constraint conditions to construct compact FP-tree and mine inter-transactional association rules. It is presented that the main idea and the pseudo-code of ITARM. A performance test is done for the algorithm, the results show that ITARM is an algorithm with high temporal and spatial performances.