Features in recognizing activities of daily life in smart homes

Abstract

The question arises of how to ensure a safe stay for the elderly in their home environment. Development in sensor technology, machine learning, and artificial intelligence can contribute to achieving this goal. Activities of daily life recognition systems recognize the activities of residents in smart homes with high accuracy. They can consequently detect anomalies in the daily functioning of the elderly, which may indicate health problems. Systems of this type are based on sensor networks and recognize the activities of residents based on the activation of various sensors. The accuracy of activity recognition depends primarily on the quality of the features that are compiled from sensor data and the classification model, which is built from the collection of data in the learning phase. In the presented research, we mainly focus on selecting and extracting features and studying how different features affect the final result of activity recognition. We analyze which activities have the best and which have the worst recognition accuracy. In the latter's case, we also show which activities are most often misidentified.