MINING AND PREDICTING RATE-OF-RISE HEAT DETECTOR DATA
. Extreme events like fire can cause massive damage to indoor areas and life threatening conditions. Early residential fire detection is important for life preservation, prompt extinguishing and reducing damage. To detect fire, one or a combination of sensors (heat detectors, smoke detectors, flame detectors) and a detection algorithm are needed. The sensors might be a part of a wireless sensor network (WSN) or work independently. One of the most frequently used heat detectors is the rate-of-rise heat detector. In this paper some of the data mining algorithms on simulation data of the rate-of-rise heat detector are applied. Data mining seems to be an effective technique for discovering useful knowledge from a large amount of data observed by many sensors. Prediction in sensor networks can be performed in the way that each sensor learns a local predictive model for the global target classes, using only its local input data. Only the predicted target class for each reading is then transmitted to the gateway or to the base station. One important class of such algorithms are predictors, which use the sensor inputs to predict some output function of interest. The purpose of the paper is to analyze different classification algorithms in the case of rate-of-rise heat detector to see which of the applied techniques led to higher accuracy and fewer errors.