ContextAct @ A4H Dataset

Published: 19-02-2019| Version 3 | DOI: 10.17632/fcj2hmz5kb.3
Contributor:
Paula Lago

Description

ContextAct@A4H is a real-life dataset of Activities of Daily Living. ContextAct@A4H features rich, real-life daily living dataset collected in the Amiqual4Home smart apartment. It contains data from sensors in the apartment, collected while a person was living there during two periods: one in June and one in November (summer and fall respectively). Example sensor include: door sensors, electric consumption sensors (per device), light on/off, lighting levels per room, in-door location, There are a total of 9 labeled activity classes and 397 activity instances. The README file contains more details. The experiment for collecting ContextAct@A4H was performed in the frame of a collaboration between LIG, Amiqual4Home and Universidad de Los Andes (Colombia). The main contribution of this dataset is the inclusion of context variables (weather, temperature, noise, humidity, the existence of visitors, etc) and the high number of properties measured in the apartment. The dataset was presented in the CONTEXT Conference. Its first use is to perform context-aware routine learning in the frame of Ambient Assisted Living research. We envision many other uses for this dataset, including the following: * To test different sensor configurations and compare which gives best classification results for activity recognition or to test which sensors suit best your application use case, finding a trade-off between cost and accuracy. * To evaluate reliability in sensor networks since some properties have been measured using different sensors. * For energy consumption analysis and prediction using energy-related variables. * For evaluating context-aware services. For example, a context-aware music recommendation using music information available in the dataset. * For routine learning and analysis since the dataset is a real-life routine. Please contact Paula Lago for any questions Paula Lago pa.lago52@uniandes.edu.co

Files