ECSS Congress Search Engine

Abstract details
Abstract-ID: 723
Session: [OP-AP21] SPORT TECHNOLOGY/EQUIPMENT / ,
Lecture room:
Date & time: 03.07.2024 / -
Title of the paper: PRIVACY-PRESERVING FEDERATED LEARNING FOR ATHLETIC ACTIVITY RECOGNITION USING MOBILE SENSORS DATA
Authors: ABDUL WASAY SARDAR, CONNOR, M. , O’NEILL, M. - [Contact]
Institution: UNIVERSITY COLLEGE OF DUBLIN
Department: UCD SMURFIT SCHOOL OF BUSINESS
Country:
Abstract text

INTRODUCTION:
Athletic activity recognition is the procedure of recognizing individual or group-specific activities utilizing mobile and wearable sensors. In previous research, human activity recognition is done by applying traditional machine learning models to recognize individual activities, however, this approach can result in data security and privacy issues. To address these issues, we investigate the use of privacy-preserving federated learning models to recognize users activities without sharing data [1].
METHODS:
We design a privacy-preserving federated learning method that can recognize athletic activities by sharing training parameters not data. Our privacy-preserving federated learning model contains three steps: training the local model, sharing parameters with the global model, aggregating weights with other clients, getting back the global model, and retraining again by using the updated local model. We performed multiple rounds until we achieved maximum accuracy. Comparing the performance of each client in a federated learning approach and traditional approach. Also, implementing multiple deep learning algorithms and comparing their performance as well.
RESULTS:
The privacy-preserved federated learning approach using the LSTM model, achieved an 88.38 % average accuracy for 15 different clients, and the traditional approach 92.7%. In the case of the RNN model average testing accuracy is 88.69% and for the traditional approach testing accuracy is about 93.4%.
CONCLUSION:
Federated learning can provide a reasonable alternative to traditional approaches for activity recognition using mobile sensor data. The federated learning approach presented in this work can protect users’ privacy whilst maintaining acceptable accuracy.

1. Sozinov, K., Vlassov, V., & Girdzijauskas, S. (2018, December). Human activity recognition using federated learning. In 2018 IEEE Intl Conf on Parallel & Distributed Processing with Applications, Ubiquitous Computing & Communications, Big Data & Cloud Computing, Social Computing & Networking, Sustainable Computing & Communications (ISPA/IUCC/BDCloud/SocialCom/SustainCom) (pp. 1103-1111). IEEE.
2. Shoaib, M., Scholten, H., & Havinga, P. J. (2013, December). Towards physical activity recognition using smartphone sensors. In 2013 IEEE 10th international conference on ubiquitous intelligence and computing and 2013 IEEE 10th international conference on autonomic and trusted computing (pp. 80-87). IEEE.
3. Sardar, A. W., Ullah, F., Bacha, J., Khan, J., Ali, F., & Lee, S. (2022). Mobile sensors based platform of Human Physical Activities Recognition for COVID-19 spread minimization. Computers in Biology and Medicine, 146, 105662.
4. Chai, Y., Liu, H., Zhu, H., Pan, Y., Zhou, A., Liu, H., ... & Qian, Y. (2024). A profile similarity-based personalized federated learning method for wearable sensor-based human activity recognition. Information & Management, 103922.
5. https://www.cis.fordham.edu/wisdm/index.php

Topic: SPORT TECHNOLOGY
Keyword I:

ADODB.Recordset Fehler "800a0cc1'

Ein Objekt, das dem angeforderten Namen oder dem Ordinalverweis entspricht, kann nicht gefunden werden.

/asp/congress/myCPB0_4_Read_Account.asp, line 207