Enhancing elderly activity recognition and safety through a hybrid deep learning model

Abstract

In recent years, there has been a significant increase in interest in human activity recognition (HAR), primarily driven by the development of sensor-based technologies and their applications in various fields, including security, healthcare, and personal fitness. HAR systems have been the subject of numerous studies, most of which have concentrated on identifying everyday human activities. There is still a need to address the unique requirements of senior citizens, whose physical activity patterns vary due to age-related factors. Challenges with accuracy, flexibility of data collection, and dataset restrictions (e.g., few classifications and small sample sizes) emerge in the particular geriatric HAR setting. These problems hinder the development of reliable systems that accurately identify and track the activities of the elderly. In this research, a new approach to Elderly Activity Recognition is presented, based on a self-made Sensor-Enabled Android App that records movement features to produce a complete dataset with six different classes. We propose a hybrid model that combines Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) to address the issues mentioned above, thereby offering improved accuracy and adaptability. With a high accuracy of 98.60%, our model outperformed earlier methods by a wide margin. The model performed well when we assessed it using the following metrics: area under the curve (AUC), recall, F-score, and precision. The precision, recall, and f-score values are accordingly 98.90%, 96.79%, and 96.12%. The study’s findings offer valuable insights for developing systems that effectively identify and track the activities of the elderly, thereby enhancing their safety and overall well-being.Although the model is designed for elderly activity detection, it can be applied to a broader range of applications, including general human activity recognition, fitness tracking, rehabilitation monitoring, and fall detection.