Variational autoencoder-based anomaly detection in time series data for inventory record inaccuracy

Retail companies monitor inventory stock levels regularly and manage them based on forecasted sales to sustain their market position. Inventory accuracy, defined as the difference between warehouse records actual inventory, is critical for preventing stockouts shortages. The root causes of inaccuracy are employee or customer theft, product damage spoilage, wrong shipments. In this paper, we aim at detecting inaccurate stocks one Turkey's largest supermarket chain using variational autoencoder (VAE), which an unsupervised learning method. Based findings, showed that VAE able model underlying probability distribution data, regenerate pattern from time series detect anomalies. Hence, it reduces effort manually label in data. Since data depends selected product/product categories, had use a parametric approach handle potential differences. For individual products, built univariate series, whereas categories multivariate series. experimental results show proposed approaches can anomalies both low high quantities.