Daily Activity Patterns Recognition

• Autor(a) principal: Mário Filipe Araújo Ferreira
• Data de Publicação: 2017
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: https://hdl.handle.net/10216/106483
• topic: Engenharia electrotécnica, electrónica e informática; Electrical engineering, Electronic engineering, Information engineering

Embedding computational capability into quotidian objects is called ubiquitous computing and is a big trend in software engineering. These objects become capable of tracking their user's actions using various sensors that can collect many types of information, such as heart rate, location and movement. All this information can be used in many ways and one of them is to compare past and present sensor metrics to establish connections and similarities between activities, with the final objective of gaining knowledge about the user of the pervasive system. Fraunhofer AICOS solutions in the field of monitoring physical activity and related metrics store substantial amounts of data generated during the use of such applications by the users. The analysis of this data would lead to the understanding of user's behaviour patterns, that could greatly improve the reach of the solutions already created. The lack of knowledge about the user's behaviour patterns prevents the solutions previously mentioned from personalizing their suggestions, strategies and recommendations to improve their efficiency when applied to intervene in the user's physical, cognitive or behaviour decline. It is quite important for caregivers to have a complete awareness of the user's routines so they can correct the unhealthy ones. In this work, the problem of using computational devices and algorithms to detect user's routines is addressed by using frequent pattern mining techniques. To implement these methods the data provided by the Smart Companion's applications was parsed and used as input in a set of potentially useful algorithms for frequent item set mining, like the FP-Growth and Split and Merge algorithms. The algorithm capable of producing the best results was embedded in the pattern recognition web tool which then uses visualization methods specially developed to meet user's needs. The final product was validated by potential end-users, informal caregivers of the Smart Companion's users, who utilized the pattern recognition tool to evaluate if new knowledge about the user's routine was created and appropriately conveyed.