A machine learning approach for indirect human presence detection using IOT devices
Authors:
- Rui MadeiraRMRui Madeira
- Luis NunesLNLuis Nunes
- 1 September 2016
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 145-150
Abstract
This paper describes the construction of a system that uses information from several home automation devices, to detect the presence of a person in the space where the devices are located. The detection however doesn't rely on the information of devices that explicitly detect human presence, like motion detectors or smart cameras. The information used is the one available in the Muzzley system, which is a mobile application that allows the monitoring and control of several types of devices from a single program. The provided information was anonymized at the source. The first step was to extract adequate features for this problem. A labeling step is introduced using a combination of heuristics to assert the likelihood of anyone being home at a given time, based on all information available, including, but not limited to, direct presence detectors. The solution rests mainly on the use of supervised learning algorithms to train models that detect the presence without any information based on direct presence detectors. The model should be able to detect patterns of usage when the owner is at home rather than rely only on direct sensors. Results show that detection in this context is difficult, but we believe these results shed some light on possible paths to improve the system's accuracy.Keywords
This publication has 8 references indexed in Scilit:
- The WEKA data mining softwareACM SIGKDD Explorations Newsletter, 2009
- Ambient intelligence: Technologies, applications, and opportunitiesPervasive and Mobile Computing, 2009
- How smart are our environments? An updated look at the state of the artPervasive and Mobile Computing, 2007
- True Visions2006
- The Gator Tech Smart House: a programmable pervasive spaceComputer, 2005
- The Internet of ThingsScientific American, 2004
- Random decision forests2002
- A training algorithm for optimal margin classifiers1992