<![CDATA[This study develops an early warning system for detecting operational anomalies (red flags) in the food and beverage industry, specifically for businesses lacking historical fraud data, using Terminal Coffee as a case study. The system integrates rule-based methods and logistic regression, combining classification logic and probabilistic prediction. The initial stage applies a rule-based approach by setting statistical thresholds (mean ± standard deviation) for eight operational indicators, including COGS ratio, electricity cost, average items and sales per transaction, and discount-to-sales ratio. From a total of 1,449 shift-level observations collected over one year, 436 (30.09%) were classified as red flags. These classifications were then used as the dependent variable in a binary logistic regression model. The estimation results identified four statistically significant predictors (p < 0.05): COGS per item, average sales per transaction, average sales per item, and discount ratio. The final model demonstrated strong classification performance, with 92% accuracy, 83% sensitivity, 95.3% specificity, 86.7% precision, and an AUC of 0.957—indicating excellent discriminatory ability. These findings suggest that combining rule-based logic and logistic regression effectively builds a reliable and adaptive early warning system for operational monitoring, even in the absence of historical fraud records. The proposed system is applicable for integration into managerial dashboards as a data-driven decision support tool to facilitate proactive, objective, and timely interventions in daily operational oversight. Key Words : red flags, rule-based, logistic regression, anomaly detection, early warning system.]]>
