Fall Detection Using Federated Learning For Model Personalization And Anomaly Detection
dc.contributor.advisor | Ngu, Anne Hee Hiong | |
dc.contributor.author | Maray, Nader | |
dc.contributor.committeeMember | Metsis, Vangelis | |
dc.contributor.committeeMember | Lee, Chul-Ho | |
dc.date.accessioned | 2023-04-21T18:22:26Z | |
dc.date.available | 2023-04-21T18:22:26Z | |
dc.date.issued | 2023-04 | |
dc.description.abstract | Falls in the elderly are associated with significant morbidity and mortality. While numerous fall detection devices incorporating AI and machine learning algorithms have been developed, no known smartwatch-based system has been used successfully in real-time to detect falls for elderly persons. We have developed and deployed a SmartFall system on a commodity-based smartwatch which has been trialled by nine elderly participants. The system, while being usable and welcomed by the participants in our trials, has two serious limitations. The first limitation is the inability to collect a large amount of personalized data for training. When the fall detection model, which is trained with insufficient data, is used in the real world, it generates a large amount of false positives. The second limitation is the model drift problem. This means an accurate model trained using data collected with a specific device performs sub-par when used in another device. Therefore, building one model for each type of device/watch is not a scalable approach for developing smartwatch-based fall detection system. To tackle those issues, we will focus on two datasets including accelerometer data for fall detection problem from different devices: the Microsoft watch (MSBAND), and the Meta Sensor device. We have previously achieved good success in solving the limitations through the use of transfer learning, however, false positives still remained as a problem, as well as real-time model testing. To solve the remaining issues, we will try to apply two main methods, the first method is building a federated learning framework of multiple edge devices for multiple people, where each edge device would have its own personalized fall detection model, as well as its own synthetically generated data and real life data, while the second method would treat the problem as an anomaly detection problem, where the standard action would be an ADL (Activities of Daily Life) and the anomaly would be a fall, and vice versa. Federated learning experiments showed promising results, achieving an average F1-score of 0.94, while the anomaly detection experiment did not achieve results better than the previous model. | |
dc.description.department | Computer Science | |
dc.format | Text | |
dc.format.extent | 51 pages | |
dc.format.medium | 1 file (.pdf) | |
dc.identifier.citation | Maray, N. (2023). Fall detection using federated learning for model personalization and anomaly detection (Unpublished thesis). Texas State University, San Marcos, Texas. | |
dc.identifier.uri | https://hdl.handle.net/10877/16627 | |
dc.language.iso | en | |
dc.subject | fall detection | |
dc.subject | model personalization | |
dc.subject | machine learning | |
dc.subject | deep learning | |
dc.subject | LSTM | |
dc.subject | recurrent neural networks | |
dc.subject | Smartwatch | |
dc.subject | accelerometer | |
dc.title | Fall Detection Using Federated Learning For Model Personalization And Anomaly Detection | |
dc.type | Thesis | |
thesis.degree.department | Computer Science | |
thesis.degree.discipline | Computer Science | |
thesis.degree.grantor | Texas State University | |
thesis.degree.level | Masters | |
thesis.degree.name | Master of Science |
Files
Original bundle
1 - 1 of 1