Real-time Monitoring of Activity Recognition in Smart Homes: An Intelligent IoT Framework

Ahmed Aziz1,*, Sanjar Mirzaliev1, Yuldashev Maqsudjon1

1Tashkent State Universtiy of Economics, Tashkent, Uzbekistan

Emails: a.mohamed@tsue.uz; sanjar2611@gmail.com; maqsudjon.yuldashev@tsue.uz

Text Box: Abstract The rapid proliferation of the Internet of Things (IoT) has paved the way for transformative innovations, and this paper explores its profound impact on the realm of elderly care within smart homes. We present a pioneering IoT-based approach for human activity recognition, addressing the critical need for accurate and non-intrusive monitoring of elderly individuals. Our IoT-based approach begins with data preprocessing, where raw sensor data is refined using median filtering, reducing noise and ensuring high-quality inputs for our model. We apply the "series_to_supervised" transformation to convert the sensor data into a supervised learning format, which is critical for training the GRU-based activity recognition model. The heart of our approach lies in the federated distillation-based training strategy. Edge devices within the IoT network locally train their GRU models using their datasets while sharing knowledge with a central server and other edge devices. Knowledge distillation further enhances the model's performance by transferring knowledge from the global model to the edge devices. Experimental analysis demonstrated an impressive accuracy of 95% and an F1-score of 0.94, Our system excels in recognizing and classifying a wide range of human activities, from daily routines to emergencies. *Corresponding Author: a.mohamed@tsue.uz

Received: March 28, 2023  Revised: June 22, 2023  Accepted: September 13, 2023

Keywords:     IoT (Internet of Things); Elderly Care;  Smart Homes; Human Activity Recognition;    Ambient Assisted Living (AAL); Sensor Networks;   Healthcare Applications;   Wearable Devices.