UseofAITechniquesfor.PDFVIP

Use of AI Techniques for Residential Fire Detection in Wireless Sensor Networks Majid Bahrepour, Nirvana Meratnia, Paul J. M. Havinga (m.bahrepour, n.meratnia, p.j.m.havinga)@utwente.nl Pervasive Systems Research Group, Twente University, the Netherlands Abstract: Early residential fire detection is important for prompt extinguishing and reducing damages and life losses. To detect fire, one or a combination of sen- sors and a detection algorithm are needed. The sensors might be part of a wireless sensor network (WSN) or work independently. The previous research in the area of fire detection using WSN has paid little or no attention to investigate the optim- al set of sensors as well as use of learning mechanisms and Artificial Intelligence (AI) techniques. They have only made some assumptions on what might be consi- dered as appropriate sensor or an arbitrary AI technique has been used. By closing the gap between traditional fire detection techniques and modern wireless sensor network capabilities, in this paper we present a guideline on choosing the most op- timal sensor combinations for accurate residential fire detection. Additionally, ap- plicability of a feed forward neural network (FFNN) and Na?ve Bayes Classifier is investigated and results in terms of detection rate and computational complexity are analyzed. 1 Introduction Fires may take place in various environments, such as residential places, forests or open spaces. The easiest way to detect a fire at residential places is using the smoke detectors or any other similar sensors, which are usually sensitive to ioniza- tion or obscuration [1]. The problem with such detectors is that they