Privacy-Preserving Fall Detection using Federated Machine Learning in IoT-based Applications
(2025) EITM01 20251Department of Electrical and Information Technology
- Abstract
- Falls are one of the leading causes of injury and death among the elderly, creating a growing need for efficient and privacy-preserving fall detection systems. Traditional machine learning approaches for fall detection often rely on centralized data collection, where sensitive user data is collected and stored on a central server. This poses significant privacy risks, especially in the healthcare sector.
This work explores the possibility of using federated machine learning for fall detection as an alternative to traditional centralized machine learning. Federated learning enables model training directly on user devices, such as smartphones and wearable sensors, without the need to transfer raw data to a central server. The thesis... (More) - Falls are one of the leading causes of injury and death among the elderly, creating a growing need for efficient and privacy-preserving fall detection systems. Traditional machine learning approaches for fall detection often rely on centralized data collection, where sensitive user data is collected and stored on a central server. This poses significant privacy risks, especially in the healthcare sector.
This work explores the possibility of using federated machine learning for fall detection as an alternative to traditional centralized machine learning. Federated learning enables model training directly on user devices, such as smartphones and wearable sensors, without the need to transfer raw data to a central server. The thesis compares the performance and privacy protection of centralized and federated learning through experiments based on the publicly available MobiAct dataset.
The results show that federated machine learning can reduce privacy risks while maintaining competitive accuracy compared to a centralized model. Furthermore, potential challenges and opportunities of integrating federated learning into Internet-of-Things-based systems are discussed. The study thus contributes insights into how federated machine learning can be used to create more secure and privacy-preserving solutions in this area. (Less) - Popular Abstract
- Fall incidents are one of the most common causes of injuries and fatalities among the elderly. Modern technology can help detect falls, but this often comes at the expense of personal privacy. In our thesis, we explore how federated machine learning can be used to detect falls without sharing any raw motion data from users’ devices, thereby protecting personal privacy.
Each year, millions of elderly individuals suffer from fall incidents, leading to extensive hospital stays and, in the worst cases, fatal outcomes. Modern technologies can assist in detecting and responding to fall incidents more quickly, but many of these systems are based on centralized machine learning, where all data is collected and analyzed on a central server. This... (More) - Fall incidents are one of the most common causes of injuries and fatalities among the elderly. Modern technology can help detect falls, but this often comes at the expense of personal privacy. In our thesis, we explore how federated machine learning can be used to detect falls without sharing any raw motion data from users’ devices, thereby protecting personal privacy.
Each year, millions of elderly individuals suffer from fall incidents, leading to extensive hospital stays and, in the worst cases, fatal outcomes. Modern technologies can assist in detecting and responding to fall incidents more quickly, but many of these systems are based on centralized machine learning, where all data is collected and analyzed on a central server. This poses a significant privacy risk, as sensitive health data can be exposed to leaks or misuse.
In our thesis, we investigate how the federated machine learning approach can provide a more secure alternative for fall detection. Federated learning allows the ML model to be trained directly on the users’ own devices, such as smartphones and wearable sensors, without requiring any sensitive motion data to be sent to a central server. Instead, only model updates are sent and aggregated to improve the system’s ability to detect falls.
To assess how this method performs in practice, we conducted an experimental study using the publicly available MobiAct dataset, which contains motion data from both daily activities and falls. We compared the performance of a federated model with that of a traditional centralized model and found that the federated solution offers nearly the same level of accuracy while significantly reducing privacy risks.
The results of our study show that federated machine learning can be an effective way to protect user privacy without compromising model performance in fall detection. This insight is crucial for the future development of health technology, particularly in elderly care and personal health monitoring where data often can be seen as very privacy sensitive. The next step in research could be to test the model in real-world environments and explore how it can be integrated into existing IoT-based healthcare platforms.
By combining advanced technology with secure data management, we can create the next generation of fall detection systems that are both effective and safe for its users. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9187822
- author
- Svegerud, Axel LU and Dahlström, Björn LU
- supervisor
-
- Kaan Bür LU
- organization
- course
- EITM01 20251
- year
- 2025
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Fall Detection, Federated Learning, Machine Learning, Data Privacy, Privacy Preservation, IoT, Sensor Data
- report number
- LU/LTH-EIT 2025-1047
- language
- English
- id
- 9187822
- date added to LUP
- 2025-05-06 10:00:21
- date last changed
- 2025-05-06 10:00:21
@misc{9187822,
abstract = {{Falls are one of the leading causes of injury and death among the elderly, creating a growing need for efficient and privacy-preserving fall detection systems. Traditional machine learning approaches for fall detection often rely on centralized data collection, where sensitive user data is collected and stored on a central server. This poses significant privacy risks, especially in the healthcare sector.
This work explores the possibility of using federated machine learning for fall detection as an alternative to traditional centralized machine learning. Federated learning enables model training directly on user devices, such as smartphones and wearable sensors, without the need to transfer raw data to a central server. The thesis compares the performance and privacy protection of centralized and federated learning through experiments based on the publicly available MobiAct dataset.
The results show that federated machine learning can reduce privacy risks while maintaining competitive accuracy compared to a centralized model. Furthermore, potential challenges and opportunities of integrating federated learning into Internet-of-Things-based systems are discussed. The study thus contributes insights into how federated machine learning can be used to create more secure and privacy-preserving solutions in this area.}},
author = {{Svegerud, Axel and Dahlström, Björn}},
language = {{eng}},
note = {{Student Paper}},
title = {{Privacy-Preserving Fall Detection using Federated Machine Learning in IoT-based Applications}},
year = {{2025}},
}