Research on the Model of Anomaly Detection of FMCG Based on Time Series Illustrated by the Case of Cosmetics

摘要

FMCG (Fast Moving Consumer Goods) as one of the most attractive commodities in the logistics industry, its small number of consumption patterns affects all aspects of supply chain activities. FMCG is an impulse buying product, which will affect the uncertainty of demand and the bullwhip effect caused by demand fluctuation. According to the characteristics of cosmetics, this paper divides the products into two categories, then designs and improves two abnormal detection algorithms to avoid the impact of impulse consumption on demand. Aiming at a large number of common commodities, the K-iForest algorithm based on Isolation Forest (iForest) algorithm and K-means algorithm is proposed to control the sensitivity of the algorithm while ensuring the efficiency of the algorithm. For the luxury goods, the Support Vector Regression (SVR) model with sliding window is improved, which makes the algorithm more inclusive for small fluctuations.