PRIVACY-PRESERVING HUMAN ACTIVITY RECOGNITION USING LIGHTWEIGHT MACHINE LEARNING ALGORITHMS AND MULTI-SOURCE DATASETS

---

Tezin Türü: Yüksek Lisans

Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Odtü Kuzey Kıbrıs Kampusu (Kktc-Güzelyurt), Bilgisayar Mühendisliği Bölümü, Türkiye

Tezin Dili: İngilizce

Öğrenci: Adeena Toaha

Danışman: Enver Ever

Özet:

With the growing popularity of wearable devices and the increasing demand for data-driven healthcare applications, there has been a surge in the development of human activity recognition (HAR) systems. However, many current HAR systems rely on computationally intensive models that require data to be sent to remote servers, posing privacy concerns. This study proposes an investigation of privacy-preserving HAR systems that utilize lightweight machine learning algorithms and leverage information from multi-source datasets to address this issue.

The proposed system employs lightweight algorithms specifically designed for edge devices and enables efficient processing of data specifically designed for edge devices and efficient data processing locally. Such an approach is expected to come with a degree of user privacy by not sharing sensitive personal data while achieving accuracy comparable to or superior to existing methods.

The systems performance is further enhanced by incorporating data from publicly available HAR datasets such as MHealth, OPPORTUNITY, and OPPORTUNITY++. These datasets comprehensively represent human activities and environments, allowing the system to learn from a wider range of data and improve its generalizability.

The proposed system offers a promising solution for privacy-preserving HAR on edge devices by combining lightweight machine learning algorithms with insights from multi-source datasets. This approach can pave the way for the development of personalized and ubiquitous healthcare applications that balance accuracy and privacy concerns.