Research on the Application of Statistical Anomaly Detection in Tobacco Sales Anomaly Detection

Abstract

With the rapid development of a new generation of information technology, digitalization and informatization have become key factors in enhancing the core competitiveness of the tobacco industry. Detecting and monitoring abnormal tobacco sales behaviors play a vital role in combating illegal transactions and ensuring the healthy development of the tobacco industry. Manual identification of abnormal sales behaviors presents practical issues, such as low efficiency, strong subjectivity, and a reliance on accumulated experience. Utilizing statistical probability theory to help the tobacco bureau determine the threshold for defining abnormal sales behavior can effectively improve the efficiency of anomaly detection. This study applies relevant theories of statistical anomaly detection in combination with practical production needs, establishing four threshold criteria for identifying abnormal sales behaviors: the number of days without retail data, the ratio of daily sales volume to the sales volume of the previous seven days, the number of days of available sales, and the number of days with scanning activity per week. These criteria make it possible to automate and dynamically identify abnormal sales behaviors in subsequent sales situations.