Hybrid Convolutional Neural Network Ensemble for Activity Recognition in Mobile Phones

The increasing importance of human activity recognition in Ambient Assisted Living systems (AAL) has brought a tremendous growth in research efforts in the field. Amongst these, the identification of activities of daily living using sensors available in mobile devices has emerged as one of the most interesting goals for AAL systems. With the development of deep learning algorithms in recent years, there has been increasing interest in hybrid models, which are able to perform as well as traditional machine learning models and yet entail the self-learning capabilities of deep learning at the same time. Using the time series data obtained from the MotionSense dataset, this paper introduces the concept of converting time series signals to images. The paper then proposes a novel approach for activity recognition by fusing a variety of traditional machine learning models with a deep convolutional neural network via a majority voting ensemble. The proposed method produced better results than the traditional machine learning models.