Utility and occupancy driven pattern analysis for processing dynamic data streams in damped window control

Abstract

Data analysis is suitable for data control systems by discovering hidden knowledge that is difficult for humans to perceive from huge and complex data. In various data analysis methods, high utility occupancy pattern analysis considers the utility occupancy of each pattern in the corresponding transaction in addition to the profit and quantity of patterns, which is effective for data control systems, including data science fields. However, recent data holds more insightful knowledge when processing real-time generated data. Previous occupancy-based approaches do not handle the relative significance of the latest data. To overcome the limitation, we introduce a new method for discovering high utility occupancy patterns from dynamic data streams where time-sensitive data consistently occurs. The proposed method assigns relative importance to each pattern by considering the temporal aspect of each transaction. Advanced constructing and restructuring processes are utilized in the proposed method for efficiently controlling data according to the time flow of each pattern in dynamic environments. In the pattern expansion process, a new upper bound adopting the decaying factor is suggested to efficiently reduce unnecessary searches for unpromising patterns. Experimental results demonstrate that the proposed method has superior runtime and scalability performance compared to state-of-the-art methods with comparable memory usage. The ablation study underscores how the proposed components contribute to the overall effectiveness of the proposed method. Additional evaluations indicate that the proposed method analyzes insightful result patterns compared to state-of-the-art methods, and a case study demonstrates its applicability to real-time dynamic data control systems.