Trend-Aware Rule Mining: A Lightweight Framework for Data-Scarce Retail Decision-Making

Small and medium-sized enterprises (SMEs), particularly in the retail and food sectors, often face urgent decisions such as product bundling or menu design with only a few weeks of sales data. Traditional Association Rule Mining (ARM) techniques are not well-suited for this context, as they rely on large, stable datasets and ignore short-term behavioral shifts. This paper proposes a lightweight and interpretable framework called Trend-Aware Rule Mining (TARM), designed specifically for data-scarce, early-stage retail environments. Using only three months of point-of-sale transaction data from a newly opened coffee shop, TARM integrates standard confidence scoring with time-slot-based trend adjustments. The method calculates average weekly confidence, detects relative frequency trends, and adjusts rule strength using a bounded nonlinear function. Results show that TARM successfully highlights item pairs that are not only frequently purchased together but are also gaining relevance within specific time windows, enabling time-sensitive decisions such as bundling and targeted promotions.