Enhancing Elderly Safety: A Machine Learning-Driven Acoustic System for In-Home Fall Detection and Alerting

The rapidly aging global population and the increase in time spent at home by the elderly, exacerbated by the COVID-19 pandemic, have highlighted a critical need for innovative healthcare solutions. One of the most pressing concerns in elderly care is the high risk and frequency of falls, which are the leading cause of accidental deaths and severe injuries in this demographic. With current technologies largely relying on visual systems that raise privacy concerns and incur significant costs, there is an urgent need for a more practical and respectful solution. This paper presents a novel machine learning-based approach to detect falls among the elderly using sound as the primary sensory input. We have developed an intelligent terminal that employs sound sensors to collect audio information within home settings, thereby creating a comprehensive database of real elderly fall incidents. By extracting kinematic, temporal, and dynamic features from actual falls and everyday activities, we designed and trained machine learning algorithms to discern fall events with high accuracy. Comparative assessments of the developed algorithms against benchmark classification metrics, such as the Area Under the Curve (AUC), reveal the robustness of our approach, with the best-performing models achieving AUC values exceeding 0.95. The system proposed herein offers a low-cost, privacy-preserving, and low-latency alternative to camera-based fall detection systems, showing great promise for widespread adoption in aging societies and private homes. This research signifies a step forward in facilitating safer independent living conditions for the elderly and reducing the healthcare burden associated with falls.