A knowledge-driven approach for activity recognition in smart homes based on activity profiling

The Internet of Things (IoT) is a technology for seamlessly connecting a large number of small-end devices and enabling the development of many smart applications to control different aspects of our life; shifting us, ever-closer to living in a smart city. IoT makes it possible to convert our homes to smart environments in which sensors are responsible for handling inhabitants' behaviours and monitor their daily activities. Activity Recognition (AR) is a new service within smart homes. It has been introduced as a solution to improve the quality of life of people such as elderly and children. AR is concerned with the assignment of an activity label to a sequence of sensors' events that are generated from the smart infrastructure. To help in effectively recognizing home activities, classification algorithms are applied on segmented sequences that are extracted automatically. Segments are subject to error due to the existence of irrelevant data and difficulties in how segmentation is applied. This negatively affects the accuracy on the classification task. In addition, the data generated from the network is streamed in nature, and big data techniques need to be utilized. In this paper, we propose a model to improve Activity Recognition in smart homes. The proposed technique is based on defining a profile for each activity from training datasets. The profile will be used to induce extra features and will help in distinguishing residents' activities (fingerprinting). To validate our model, real datasets have been used for the experiments, and results show a significant enhancement in accuracy, compared with traditional techniques. (c) 2017 Elsevier B.V. All rights reserved.