ELDERLY FALL DETECTION USING UNSUPERVISED TRANSFORMER MODEL
Іван Урсул · Electronics and Information Technologies · 2024
This study explores the development of an unsupervised Transformer model for fall detection among the elderly. It aims to help address the critical need for reliable fall detection systems in an aging global population. Traditional methods, limited by accuracy and privacy concerns, require innovative approaches. The research employs unsupervised learning techniques to analyze accelerometer data, aiming to enhance detection results without compromising privacy. The Transformer model’s performance, assessed through Mean Squared Error and Root Mean Squared Error, demonstrates a high degree of efficacy in reconstructing accelerometer data, which is crucial for identifying falls as anomalies. Results indicate that the model’s precision is very close to supervised models, with MSE and RMSE values suggesting significant accuracy. This approach reduces the reliance on extensive labeled datasets, a common challenge in fall detection research. Hence, it offers a practical solution in real-world scenarios where labeled data is scarce. The study’s findings underscore the potential of unsupervised learning models in advancing fall detection technologies. This is promising in improving healthcare outcomes for the elderly and paving the way for broader applications in activity monitoring and anomaly detection. Keywords : Fall Detection, Unsupervised Learning, Transformer Model, Accelerometer Data, Elderly Care, Anomaly Detection, Geriatric Healthcare.