Radio Frequency Identification (RFID) technology enables high-speed identification and tracking of physical objects without line-of-sight, making it ideal for applications in logistics and inventory management.
While RFID is earning a substantial momentum as a preferred solution for automatic identification and data collection, it still presents several challenges. RFID systems typically generate large volumes of data, leading to issues such as missed readings, duplicate entries, noisy data, and long processing times—particularly under varying tag arrival and noise rates. These challenges may prevent the widespread adoption of RFID-based systems.
RFID data cannot be directly used by applications unless it is properly filtered and cleaned. Although RFID systems generate data at high rates and in large volumes, effective detection requires efficient processing—especially for real-time monitoring applications.
This thesis addresses these challenges through the lens of Inventory Business Intelligence (IBI), a domain that demands rapid and accurate decision-making in supply chain environments. Three key enhancements to RFID data filtering and cleaning are proposed: (1) filtering and cleaning with a fixed threshold, (2) filtering and cleaning with a dynamic threshold and (3) filtering and cleaning with a dynamic threshold combined with a reduced number of readings. These key enhancements significantly reduce data noise and processing time—achieving up to a threefold improvement over state-of-the-art techniques under various tag arrival and noise conditions.
Experimental results showed that the third enhancement achieves perfect performance, with all evaluation metrics—including accuracy, sensitivity, specificity, precision, F-score, and Matthews Correlation Coefficient (MCC)—reaching 100%. These results were obtained by comparing the third enhancement against the first and second enhancements using standard global evaluation metrics, and demonstrating its superior effectiveness in improving RFID data quality and processing efficiency.
Furthermore, system-specific metrics—including total reads, noise rate, and detection time—also confirmed the superiority of the third enhancement. It achieved the lowest total reads (1533 vs. 1732 and 2700), reduced noise to 50.19%, which is significantly lower than the first enhancement (66.41%) and comparable to the second enhancement (49.87%), and significantly minimized detection time (0.3 seconds), outperforming the first (1.66 seconds) and second (0.5 seconds) enhancements.
Finally, experimental results demonstrated that the third enhancement significantly outperforms existing techniques—namely, Dynamic RFID Data Filtering and De-noising and Duplication Elimination—in terms of both noise reduction and processing speed. At high tag arrival rates (e.g., 500 tags), it achieved 0.007 seconds, compared to 0.1 and 0.15 seconds, respectively. Additionally, it recorded lower noise rates (0.03–0.08) versus 0.09–0.11 and 0.14–0.15. These outcomes confirmed its superior performance and practical effectiveness.
