식음료 제품의 수요예측 정확도 향상을 위한 수요특성 분석과 군집화 방안

초록

This study advances demand forecasting in the food and beverage sector by profiling B2B and B2C demand via mixed-type clustering and deriving cluster-specific modeling guidelines. Integrating Coefficient of Variation (CV), Residual Strength (RS), and Product Life Cycle (PLC) for ten SKUs in each market, we identify four archetypes (Stable, Noisy, Volatile, Unstable). Late-maturity products constitute a low-variability benchmark for stable production and inventory planning. Unstable clusters favor ML-based ensembles with exogenous inputs; Volatile clusters perform best with exogenous-sensitive hybrids (e.g., SARIMAX plus tree-based learners) that capture nonlinear spikes; and Noisy clusters require robust methods with quantitative assessment of exogenous-variable effects. Academically, the study extends forecasting frameworks by systematically integrating time-series features, structural patterns, and exogenous drivers; practically, it offers actionable guidelines for managing product groups by demand type to improve forecast accuracy and strengthen supply-chain decisions. In a B2C hold-out evaluation (10 SKUs), a cluster-optimal hybrid outperformed OSFA baselines, reinforcing these design implications.

키워드