Utilizing Random Forest with iForest-Based Outlier Detection and SMOTE to Detect Movement and Direction of RFID Tags

Abstract

In recent years, radio frequency identification (RFID) technology has been utilized to monitor product movements within a supply chain in real time. By utilizing RFID technology, the products can be tracked automatically in real-time. However, the RFID cannot detect the movement and direction of the tag. This study investigates the performance of machine learning (ML) algorithms to detect the movement and direction of passive RFID tags. The dataset utilized in this study was created by considering a variety of conceivable tag motions and directions that may occur in actual warehouse settings, such as going inside and out of the gate, moving close to the gate, turning around, and static tags. The statistical features are derived from the received signal strength (RSS) and the timestamp of tags. Our proposed model combined Isolation Forest (iForest) outlier detection, Synthetic Minority Over Sampling Technique (SMOTE) and Random Forest (RF) has shown the highest accuracy up to 94.251% as compared to other ML models in detecting the movement and direction of RFID tags. In addition, we demonstrated the proposed classification model could be applied to a web-based monitoring system, so that tagged products that move in or out through a gate can be correctly identified. This study is expected to improve the RFID gate on detecting the status of products (being received or delivered) automatically.

Keywords

• RFID
• IoT
• machine learning
• tag direction
• outlier detection
• data balancing