Fuzzy Utility Stream Pattern Analysis with Temporal Aspects on Damped Window Model

Abstract Fuzzy logic provides a framework for modeling linguistic imprecision in real-world data. It has been widely applied to pattern analysis to enhance interpretability for supporting decision-making processes. High temporal fuzzy utility pattern analysis has garnered increasing attention for its ability to extract interpretable and representative patterns by integrating temporal characteristics, enabling more informed decisions in areas such as promotion planning or resource allocation. However, previous approaches in this domain operate under the assumption that all transactions hold equal significance, irrespective of their arrival times. This assumption is unrealistic, as real-world data evolve and recent transactions hold greater importance, especially for decisions that rely on timely trend responsiveness. To resolve this shortcoming, we propose an efficient approach for analyzing high temporal fuzzy utility patterns over data streams, grounded in the damped window technique. The proposed approach leverages a decaying factor to emphasize the significance of recent data. To achieve computational efficiency, the pattern expansion process employs list-based structures along with multiple pruning strategies. Comprehensive experiments indicate that the proposed method surpasses previous methods regarding runtime and scalability with competitive memory usage while exhibiting the significance of extracted results and showing practical availability through a case study. The analysis of extracted patterns highlights their relevance within a practical retail scenario, and experiments with varying decay factors corroborate the efficiency and adaptability of the proposed method for offering timely and interpretable insights in dynamic environments.