A Data-Driven Approach for Elderly Activity Recognition: Using Wearable Sensors and AI Models

As the elderly population rises in healthcare centers, monitoring their daily performance makes it difficult to identify miracle signs of an anomaly or an emergency. Activity recognition systems (ARS) to enable elderly care have become necessary because such persons require independence with consideration of their safety. Nonetheless, there is some discomfort in implementing HAR, such as sensor accuracy, a variation in health among seniors, and cognitive diversity. This paper focuses on the analysis of HAR systems for recognizing the behavior of the elderly through smartphone sensors, including accelerometers and gyroscopes. To enhance the studies of data classification, machine learning classifiers such as k-Nearest Neighbors (k-NN) and Random Forest were utilized to increase the overall accuracy. Linear regression was also employed to analyze sensor information and individualize observation. The accuracy of the chosen models indicates that the Random Forest model is best to predict the training results288. The abilities shown by these models included handling data unpredictably and featuring important information that was deemed valid. Furthermore, a possible incorporation of HAR systems with smart grid (SG) monitoring is discussed as an extension of the work for elderly care in smart homes. The feasibility of the proposed system was further demonstrated using evaluation criteria of accuracy, precision, and recall. Based on the findings of this research, we can recognize that integrating AI-driven HAR models with SG technologies might enhance elderly monitoring, creating a more innovative, accurate, and individual approach to senior care. This research provides a foundation from which to improve the living standards of elderly people, allowing them to live their remaining years safely and as independently as possible.