Training and using an inventory anomaly prediction model to predict inventory anomalies

Abstract

A computerized method trains an inventory anomaly prediction model and uses the model to predict and address inventory anomalies. Daily inventory data associated with an inventory storage location is obtained from an inventory data store and provided to a pretrained anomaly prediction model as input. The anomaly prediction model generates an anomaly indicator associated with an item type in the obtained inventory data. Corrective action is then caused to be performed in association with the item type and based on the anomaly indicator, whereby inaccuracy of an inventory count of the item type in the inventory data store is corrected. Further, in some examples, the anomaly prediction model is trained using machine learning techniques and training data that is generated and labeled using a tuned outlier detection model. Thus, the method enables the anomaly prediction model to be efficiently trained for accuracy using the described two-stage training process.