Research on abnormal inventory detection algorithm in warehouse management system

This paper proposes an anomaly detection algorithm with multi-model fusion architecture for the problem of inventory anomaly detection in e-commerce warehouse management. The algorithm combines the improved local anomaly factor algorithm with the recurrent neural network of attention mechanism, and fuses the multi-model results through Bayesian model averaging. Validated on a real e-commerce inventory dataset containing 12,358 SKUs, the detection F1 score reaches 87.3%, which outperforms existing methods. The algorithm shows strong robustness and adaptability to identify multiple types of inventory anomalies, and has been successfully applied to real production environments, significantly reducing inventory costs and improving order fulfilment rates.