<![CDATA[This research implements an automated warehouse inventory monitoring system using a hybrid approach that combines Isolation Forest and rule-based prioritization for anomaly detection, while Long Short-Term Memory (LSTM) is used for demand forecasting. The system automatically detects three types of anomalies: spike (quantity surge), drop (quantity decline), and pattern shift (changes in transaction patterns). The method utilizes 341,879 inventory records over two and a half years, combining Isolation Forest for outlier detection, rule-based prioritization for spike and drop detection, and LSTM for time series forecasting. The system is equipped with a near real-time dashboard (periodic auto-refresh), API service, and a business validation mechanism involving warehouse users. The system generated 211,450 detection candidates, consisting of 187,537 operational anomaly alerts and 23,913 INFO-level records, with accuracy of 93.0%, precision of 88.7%, recall of 100%, and F1-score of 94.0%. The INFO category is not displayed on the dashboard due to low confidence scores. The near real-time dashboard displays HIGH, MEDIUM, and LOW priority alerts with auto-refresh. Business validation achieved an 84% confirmation rate. The system also provides LSTM forecast features for predicting inventory needs for 7, 14, and 30 days ahead. This research shows that AI implementation for warehouse inventory monitoring can automate anomaly detection and improve inventory issue identification. The system shows potential for medium to large-scale warehouses requiring near real-time monitoring.]]>
Copyrights © 2026
