Incremental mining for frequent patterns in evolving time series datatabases

Several emerging applications warrant mining and discovering hidden frequent patterns in time series databases, e.g., sensor networks, environment monitoring, and inventory stock monitoring. Time series databases are characterized by two features: (1) The continuous arrival of data and (2) the time dimension. These features raise new challenges for data mining such as the need for online processing and incremental evaluation of the mining results. In this paper, we address the problem of discovering frequent patterns in databases with multiple time series. We propose an incremental technique for discovering the complete set of frequent patterns, i.e., discovering the frequent patterns over the entire time series in contrast to a sliding window over a portion of the time series. The proposed approach updates the mining results with the arrival of every new data item by considering only the items and patterns that may be affected by the newly arrived item. Our approach has the ability to discover frequent patterns that contain gaps between patterns’ items with a user-defined maximum gap size. The experimental evaluation illustrates that the proposed technique is efficient and outperforms recent sequential pattern incremental mining techniques.