Data-Efficient Elderly Activity Recognition Using Active and Ensemble Learning

Abstract

This method explores the application of sophisticated computational methods and sensor data. 
Deep learning (DL), machine learning (ML), and active learning (AL) are methods for identifying everyday activities among the elderly.Using the HAR70+ dataset, which captures activity patterns of older individuals, Seven activities—walking, shuffling, climbing stairs, standing, sitting, and lying down—as well as a subset of four essential activities—standing, sitting, walking, and lying down—are identified by the study using predictive models.Multiple ML algorithms (Random Forest, XGBoost, Logistic Regression, Three experiments were used to assess the KNN, Stochastic Gradient Descent, and DL techniques (Deep Neural Networks and LSTM).
LSTM achieved the highest accuracy of 95% for seven activities, while Random Forest reached 96% for the four-activity subset. These results highlight the effectiveness of integrating AL with ML and DL to improve activity recognition, personalize elderly care, enhance medication planning, and support overall well-being.