An Ultrasonic-based Metrological Approach for Fall Detection and User recognition using Supervised and Unsupervised Machine Learning Techniques
Ilaria Ciuffreda, Sara Casaccia, Gian Marco Revel · 2025
The rapid aging of the world’s population has increased the demand for innovative fall detection systems to ensure the safety and well-being of older people. Falls represent a major health risk, often leading to hospitalization, loss of independence, and increased caregiver burden. Traditional fall detection solutions, such as wearable sensors and camera-based systems, face challenges related to user compliance, privacy, and limited coverage areas. This study presents a novel ultrasonic (US)-based approach that combines fall detection and user identification using machine learning (ML) techniques. Unlike traditional wearable sensors, US technology offers a noninvasive and privacy-preserving alternative for monitoring falls in home and nursing home environments. The proposed system utilizes an US mounted on the ceiling for a dual purpose: fall detection and user identification. For fall detection, two ML algorithms, Random Forest and Support Vector Machine (SVM), were tested. The dataset was split using an 80/20% ratio, allocating 80% of the data for training and 20% for testing. Results indicate that Random Forest achieved high accuracy (up to 89%), demonstrating high reliability in detecting falls with minimal false negatives. Additionally, for user identification, an unsupervised K-means clustering algorithm was employed, achieving a classification accuracy of 70% based on walking patterns and height of individuals. These findings highlight the effectiveness of US-based monitoring in both detecting falls and user classification. Future work will focus on improving classification accuracy, and validating the system in real-world environments.