A Machine Learning System that Detects Abnormal Level of Inventory

Loss prevention is an important topic for retail companies to increase profitability. Losses stem from external or internal factors. In this study, we focus on an internal anomaly case about inventory movements in retail stores. Therefore, we have developed a machine learning algorithm to detect abnormal inventory levels. The machine learning algorithm developed estimates inventory levels over the last fifteen days as a time series. It then compares the actual inventory levels during this period with the predicted values to detect anomalies. This results in a store-brand level inventory error representation. A distribution of errors is obtained using errors from similar stores. The system assigns each store an anomaly label by determining whether the model-derived error is an outlier based on the distribution. Stores with labelled inventories are then sent to the business units. This allows the business units to regularly examine the stocks and predict potential fraud situations.