Enhanced human activity recognition framework for wearable devices based on explainable AI

Human Activity Recognition (HAR) has been a highly debated topic in recent years. Due to privacy concerns, especially in smart home environments, sensor-based HAR is commonly employed. This method involves users carrying smart devices, such as smartwatches or smartphones, that calculate tri-axial acceleration through gyroscopes to determine specific activities. However,Many users and developers do not fully comprehend deep learning and other algorithms. Users often worry about the unknown, and the development of HAR based on wearable devices will become increasingly challenging for developers. Consequently, the concept of Explainable AI (XAI) has been introduced to allow human users to understand and trust the machine learning algorithms and the results they produce. In this paper, we used the UCI dataset to analyze the factors that significantly influence the machine learning model with a CNN-LSTM architecture, employing XAI techniques to provide a clear demonstration of these impacts.

Text 121323.pdf - Published Version Restricted to Repository staff only Download (788kB) Official URL or Download Paper: https://ieeexplore.ieee.org/document/10791196/

Actions (login required)

View Item