Data-Driven Inventory Management for Perishable Dairy Products: Integrating Demand Segmentation, Forecast Validation, and Differentiated Replenishment Policies

Background: Inventory management for perishable products is challenging because demand variability, forecast reliability, shelf life, and batch constraints must be considered jointly. This study develops a data-driven decision-support framework for a small dairy manufacturer with limited historical inventory records. Methods: Nineteen stock-keeping units were analyzed using monthly data from September 2024 to June 2026, with July and August 2026 reserved for post-period validation. ABC analysis, demand variability, trend estimation, K-means clustering, Ward validation, and chronological forecasting models were combined with differentiated replenishment rules. Results: Four demand-pattern clusters were identified and supported by Ward clustering (adjusted Rand index = 0.8308). Random Forest achieved the best internal WAPE (27.28%), but this advantage did not persist externally, where Naive achieved 22.99%, damped Holt 26.20%, and Random Forest 30.23%. Historical demand variability was positively associated with external forecast error. Scenario analysis showed that adaptive forecast governance reduced two-month WAPE relative to maintaining a fixed Random Forest model. Conclusions: The framework supports adaptive, differentiated inventory governance by integrating economic priority, demand behavior, forecast performance, shelf life, and batch constraints without assuming unavailable optimal inventory parameters.