TIMESTAMPED ASSOCIATION RULE MINING USING USERDEFINED REFERENCE SEQUENCE:

 Abstract:

                 Given a time stamped transaction database and a user-defined reference  sequence of interest over time, Timestamped association rule mining discovers all associated item sets whose prevalence variations over time are similar to the reference sequence. The similar timestamped association patterns can expose interesting relationships of data items which co-occur with a particular event over time. Most works in timestamped association mining have focused on capturing special timestatmped regulation patterns such as cyclic patterns and calendar scheme-based patterns. However, our model is flexible in representing association patterns using a user-defined reference sequence. The dissimilarity degree of the sequence of support values of an item set to the reference sequence is used to capture how well its prevalence variation matches the reference pattern. By exploiting interesting properties such as an envelope of support time sequence and a lower bounding distance for early pruning candidate item sets, we develop an algorithm for effectively mining

timestamped association patterns.