Using unlabeled acoustic data with locality-constrained linear coding for energy-related activity recognition in buildings

Behaviors of occupants can impact on the energy consumption of buildings. Human activity recognition using smartphones as sensor platform has proliferated in recent years. With the inertial measurement unit in smartphones, behaviors of occupants in terms of walking and running could be easily identified, but the obtained information is of the simple level. Thanks to the abundant functionalities of mobile gadgets, we can now achieve a better understanding of occupants' behaviors at a more complicated level through recognition of energy-related activities by leveraging built-in microphone in smartphones. Besides, this information is not only about users themselves, but it also correlated with the appliances being used. Indeed, the recognized active activities and the associated utilized appliances will represent direct sources of energy consumption in buildings. However, many recent works on recognizing energy-related activities do require extensive labels of the dataset and the annotation process is tedious and laborious. In this paper, we aim to make use of unlabeled data to achieve satisfactory classification performance with an appropriate number of labels. In our approach, we apply the locality-constrained linear coding to process the labeled and unlabeled samples in order to achieve an acceptable classification accuracy as compared with traditional supervised learning approaches that purely rely on the large number of expensive annotations. The experimental results with both web-collected and user-recorded data show that our proposed method provides a better classification performance than the feature-engineering based supervised learning algorithms.