LIMITATIONS AND ADVANCES OF ASSOCIATION RULE MINING AND FP-GROWTH ALGORITHMS FOR TIME-VARIANT TRANSACTION DATA IN THE FOOD & BEVERAGE INDUSTRY

Abstrak

The Food & Beverage (F&B) industry exhibits high volatility, where consumer preferences and market dynamics shift rapidly due to temporal factors such as contemporary trends, seasonality, and time of day. In the era of smart restaurants, leveraging high-volume Point of Sale (POS) transaction data is crucial for optimizing supply chains and menu engineering strategies. However, conventional Association Rule Mining (ARM) algorithms, such as FP-Growth, and current predictive analytics approaches possess fundamental limitations. These methods lack native temporal modeling, rely on static threshold parameters, and are prone to high computational overhead when processing dynamic F&B transactional data characterized by frequent null transactions. Consequently, the generated patterns are often obsolete, leaving a significant gap in integrating these trend patterns with autonomous decision-making systems capable of responding to market shifts independently. To address these limitations, this study proposes a novel framework that converges Trend-Aware Data Mining, Knowledge Graphs (KG)/Ontologies, and Agentic Artificial Intelligence (Agentic AI). At the data mining layer, the FP-Growth algorithm is modified using a Time Decay Factor approach and a Sliding Window strategy to enhance sensitivity to trends and minimize runtime. The extracted temporal transactional knowledge is then integrated into a semantic ontology model and a Knowledge Graph, serving as a structured database to model interdependencies between consumer behavior and inventory dynamics. Finally, this semantic knowledge is utilized as shared global information within a Multi-Agent System (MAS) ecosystem using a Centralized-Training–Decentralized-Execution (CTDE) architecture. An agentic workflow is implemented to automate tactical decisions in supply chain management (such as automated replenishment) and dynamic menu adjustments. This research makes a significant scientific contribution through the development of an efficient, lightweight variant of the temporal FP-Growth algorithm tailored for F&B transactional data. Furthermore, it formulates a novel ontology model that bridges temporal association patterns with smart restaurant dynamics and establishes a semantic-knowledge-based Agentic AI architecture for automating supply chain and menu engineering decisions. For the industry sector, the implementation of this framework is expected to optimize stock levels, reduce food waste, and significantly enhance operational efficiency and business profitability through agile, real-time, data-driven decision-making