Abstract

Wearable sensors have become vital for health monitoring, evaluation of sport performance, and recognition of activities of daily life in Human Activity Recognition (HAR). With the combination of machine learning and inertial sensing, new possibilities for the collection of data pertaining to human motion in daily life have emerged. Nevertheless, problems remain to be resolved in the field of modeling in relation to tradeoffs in classification. The focus remains on imaging, signal variability, and sensor noise. This study aims to determine the usefulness of Inertial Measurement Unit (IMU) data for activity recognition and compares classical machine learning methods with a temporal deep learning method. IMU and Electromyography (EMG) signals in the HuGaDB dataset, a collection of multi-sensor recordings representing a set of common daily activities, were filtered, normalized, and segmented using a sliding window, and time and frequency domain features were extracted to train Logistic Regression, kNN, Random Forest, Support Vector Machine, and MLP models. An LSTM network was trained on sequential, windowed data in tandem to capture temporal dependencies of the IMU data. The findings show that classical machine learning approaches create strong baseline performance, with Random Forest achieving the highest accuracy across models that do not employ temporal methods. The outcomes reinforce the benefits of temporal modeling for HAR from wearable sensors, while also demonstrating the benefits of simpler classical modeling in terms of interpretability and lower computational requirements. The insights from this research are useful for balanced HAR design using IMU sensors and explain the trade-offs between feature-engineered ML and deep learning. The results promote HAR systems with real-world applicability by advancing multidimensional, precise, and flexible recognition systems.

Library of Congress Subject Headings

Wearable technology--Data processing; Human activity recognition; Machine learning; Deep learning (Machine learning)

Publication Date

12-2025

Document Type

Thesis

Student Type

Graduate

Degree Name

Professional Studies (MS)

Department, Program, or Center

Graduate Programs & Research

Advisor

Sanjay Modak

Advisor/Committee Member

Ayman Ibrahim

Campus

RIT Dubai

Plan Codes

PROFST-MS

Download

Share

COinS