As reported by the WHO, falls constitute the second leading cause of unintentional injury death worldwide. Particularly, adults older than 60 years suffer the most significant number of fatal falls or serious injuries, with nearly 30% of individuals over 65 reporting at least one fall annually, a risk that increases with age. The anticipated growth in life expectancy and the resulting larger aging population accentuates the economic burden associated with falls. Consequently, the identification of effective strategies for fall prevention and early detection in the elderly has become a topic of great relevance. In this study, we propose a non-invasive fall detection system based on a thermal sensor and a supervised machine-learning algorithm. The experimental dataset, generated by students through simulations of both fall and non-fall events, included the recording of room temperatures using a thermal sensor, along with the associated data labeling. For fall event detection, we evaluated three well-known supervised machine learning models: a Support Vector Machine, a Random Forest, and a Convolutional Neural Network. The experimental results demonstrate that these models exhibit robust capabilities in distinguishing between falls and non-fall events, consistently achieving performances above 95% across various evaluation metrics.

Elderly care, machine learning; sensor monitoring; fall events.