Using data mining as a tool for anomaly detection in food safety audit data. (August 2022)

• Record Type: Journal Article
• Authors: Kleboth, J.A.; Kosorus, H.; Rechberger, T.; Luning, P.A.
• Is Part Of: Food control. Volume 138(2022)
• Publication Date: 2022-08
• Subjects: Food -- Quality -- Periodicals; Food -- Analysis -- Periodicals; Food handling -- Periodicals; Food industry and trade -- Quality control -- Periodicals; Aliments -- Industrie et commerce -- Qualité -- Contrôle -- Périodiques; Aliments -- Qualité -- Périodiques; Aliments -- Analyse -- Périodiques; Hygiène alimentaire -- Périodiques; Food -- Analysis; Food handling; Food -- Quality; Periodicals; Electronic journals; 664.07
• DOI: 10.1016/j.foodcont.2022.109004
• Languages: English
• ISSNs: 0956-7135

Abstract

Abstract: The integrity of third-party food safety audits has been constantly challenged by food safety incidents of certified food businesses. Integrity programs have been established for food safety certification program owners (CPOs) to monitor the involved parties' performance and find anomalies in audit data. To find such anomalies in a large amount of data is labour intensive, and no standard approach has been established. This paper explored data mining approaches and leveraged algorithms to automate integrity checks. Furthermore, this paper provides initial validation of a suitable algorithm. Out of three potentially suitable algorithms, the couple-biased random walk (CBRW) algorithm was chosen as the basic algorithm to find anomalies in audit data. This algorithm was adjusted and expanded to show contributing factors for a potential anomaly enabling integrity managers to find the reason for potential anomalies faster. Three experts validated the sample findings of the algorithm and discussed these findings in detail. The validation showed an 80% accuracy of the algorithm and brought up findings that were not known before by the experts. The findings justify further exploration of data mining for anomaly detection in food safety audits. Highlights: Data mining for anomaly detection in audit data was investigated. Three algorithms were assessed and one algorithm was chosen. The CBRW algorithm was adapted and enhanced to gain more insights. Experts validated the findings of the algorithm in a discussion. The accuracy of the algorithm is 0, 8 and found insights the experts didn't know.