Given the ubiquity of mobile devices, understanding the context of human activity with non-intrusive solutions is of great value. A novel deep neural network model is proposed, which combines feature extraction and convolutional layers, able to recognize human physical activity in real-time from tri-axial accelerometer data when run on a mobile device. It uses a two-layer convolutional neural network to extract local features, which are combined with 40 statistical features and are fed to a fully-connected layer. It improves the classification performance, while it takes up 5–8 times less storage space and outputs more than double the throughput of the current state-of-the-art user-independent implementation on the Wireless Sensor Data Mining (WISDM) dataset. It achieves 94.18% classification accuracy on a 10-fold user-independent cross-validation of the WISDM dataset. The model is further tested on the Actitracker dataset, achieving 79.12% accuracy, while the size and throughput of the model are evaluated on a mobile device.

The critical deterioration of the environment in recent years has led to an increasing urgency of minimizing energy consumption. As residential and tertiary buildings hold a significant potential in terms of energy savings a lot of different approaches have been followed to not only reduce energy consumption but also change energy-related occupant behaviour for long-term results. Taking into account recommendation services through mobile applications, this paper proposes a more dynamic and occupant-centered solution, which not only offers useful energy related recommendations, but also adjusts the moment the recommendation is sent to maximize the receptiveness of the user in order to achieve the optimal energy behavior change. These moments are redefined micro-moments in terms of energy behavior change and occur when three factors' conditions coincide, the location, the physical and the mobile activity of the user. We develop a mobile application which can monitor them and detect a potential micro-moment, when a notification could be pushed to the user. The location is periodically collected from the device's sensors, the physical activity is inferred with the use of Google Activity Recognition API, while the mobile activity is inferred utilizing mobile event triggers and usage statistics. The experimental results show that our dynamic approach is justified, since the micro-moments are highly unpredictable and user-specific.