RADIO FREQUENCY IDENTIFICATION (RFID)-BASED ELDERLY FALL DETECTION USING LONG SHORT-TERM MEMORY (LSTM)

• 主要作者: Clairvoyance Diva P., Damianus
• 格式: Final Project
• 語言: Indonesia
• 在線閱讀: https://digilib.itb.ac.id/gdl/view/86430

Falls in the elderly can have fatal consequences if not addressed quickly, therefore it’s critical to have a system capable of reliably detecting falls and immediately notifying caregivers. Radio Frequency Identification (RFID) is one such technology. Among various configurations, the most practical setup involves positioning a reader at the edge of the room and attaching multiple passive tags to the elderly’s clothing. The way RFID works is that the reader emits a signal, which is then reflected by the RFID tags and received back by the reader. This signal strength is known as the Received Signal Strength Indicator (RSSI). Obstruction by the elderly’s body, also called body shadowing, can be used to predict indications of a fall event. In similar systems, falls are predicted using classical machine learning algorithms, though trends suggest that models based on Long Short-Term Memory (LSTM) have the potential to deliver better performance. This study shows that pure LSTM achieved 99.39% accuracy and F1macro , while CNN-LSTM achieved perfect accuracy and F1macro . These results were obtained from training and testing the model on a dataset containing 4,128 data points (1,920 positive and 2,208 negative) with 30 tags placed. When testing on continuous real- case scenarios using a sliding window approach, pure LSTM only achieved 65.63% accuracy and 65.17% F1macro , while CNN-LSTM reached 82.29% accuracy and 80.23% F1macro Additionally, this study shows that CNN-LSTM with only 10 tags in specific positions still delivered good performance, only slightly worse than pure LSTM.