Sliding window-based high utility occupancy pattern mining for data streams

High utility-based pattern mining has been proposed to analyze information by considering not only the frequency of items but also their quantity and profit. Among these, studies on high utility occupancy-based patterns have emerged, which consider the occupancy measure reflecting the share of a pattern belonging to transactions. Furthermore, as the necessity to process real-time stream data has become more critical, a method to discover high utility occupancy-based patterns in stream information has been presented recently. However, this recent method handles all accumulated data on data stream environments. Since all previously accumulated data are processed, the volume of data to be processed steadily increases over time, leading to a decline in efficiency over time. In addition, it becomes difficult to give emphasis on recent data. Consequently, these methods become less suitable for practical applications. To surmount the drawbacks, we introduce a novel approach for mining high utility occupancy patterns, employing a sliding window technique to efficiently process stream data. By focusing on fixed-size, most recent data within the window, our method effectively reflects the trends in the latest data while exhibiting improved efficiency compared to previous approaches. Extensive performance evaluations demonstrate the efficacy of the proposed method against prior methods regarding runtime, memory usage, scalability, and sensitivity. Moreover, statistical tests confirm that our approach accurately extracts the exact number of patterns without pattern loss or duplication.