A Comparative Study of Fall Detection Device for Elderly People Using Machine Learning

Abstract

Falls are a major cause of injury, disability, and hospitalization among elderly individuals, particularly those living independently without continuous supervision. The absence of timely assistance following a fall can lead to severe medical complications and increased mortality risk. To address this challenge, extensive research has been carried out in the domain of fall detection, employing a variety of approaches ranging from simple threshold-based methods to advanced Machine Learning and Deep Learning techniques. This paper presents a comparative study of existing fall detection methodologies, evaluating their accuracy, computational complexity, power consumption, and suitability for real-time wearable applications. Threshold-based systems, while computationally simple, often suffer from high false-positive rates and lack adaptability to varied movement patterns. Deep Learning-based approaches offer improved accuracy but demand significant computational resources, making them less suitable for low-power embedded devices. Based on insights drawn from this comparative analysis, the paper proposes a smart fall detection system that combines wearable sensor technology, a lightweight Machine Learning model, and IoT-based communication to achieve a practical balance between accuracy, efficiency, and cost-effectiveness. The system acquires motion data through a tri-axial accelerometer and gyroscope, applies preprocessing and feature extraction techniques, and utilizes a lightweight ML model deployed on a low-power microcontroller for real-time fall classification. Upon detecting a fall, the system triggers immediate local alerts and transmits emergency notifications along with location data to caregivers through an IoT platform, along with a reset mechanism to minimize false alarms. The comparative evaluation demonstrates that the proposed approach effectively overcomes the limitations of traditional threshold-based and resource-intensive deep learning methods, offering a scalable and reliable solution for elderly safety and remote health monitoring.