Detecting counterfeit products by means of frequent pattern mining

Product traceability is one of the major issues in supply chains management (e.g., Food, cosmetics, pharmaceutical, etc.). Several studies has shown that traceability allows targeted product recalls representing a health risk (e.g.: counterfeit products), thus enhancing the communication and risks management. It can be defined as the ability to track and trace individual items throughout their whole lifecycle from manufacturing to recycling. This includes real-time data analytics about actual product behavior (ability to track) and product historical data (ability to trace). This paper presents a comparative study between several works on product traceability and proposes a standardized traceability system architecture. In order to implement a counterfeit/nonconforming product detection algorithm, we implement a cosmetic supply chain as a multi-agent system implemented in Anylogic (c). Data generated by this simulator are then used in order to identify genuine trajectories across the whole SC. The genuine product trajectories (behavior) are inferred using a frequent pattern mining algorithm (i.e., Apriori). This identified trajectories are used as a reference in order to identify counterfeit products and detect false alarms of product behavior