Transition-Aware Human Activity Recognition using Ensemble Learning and Neural Networks

Ubiquitous computing has captivated a significant interest due to its usefulness in various domains. Human activity recognition (HAR) is an important area in ubiquitous computing which identifies human activities using sensor data. In HAR, there is a noticeable lack of focus towards transition aware HAR and most systems have focused only on identifying basic activities. However, for effective HAR, it is necessary to identify transitions accurately. In this paper, we present a novel HAR method which focuses on recognizing both basic activities and transitional activities using accelerometer and gyroscope data. We have implemented a stacked-ensemble model for activity classification using both traditional machine learning algorithms and neural networks. Our method was evaluated using a public dataset and our stacked ensemble model has achieved an overall classification accuracy of 99.74 % with oversampling. Our model accurately identifies both basic and transitional activities.