VOCs classification based on the committee of classifiers coupled with single sensor signals

A gas classification method based on a Multiple Classifiers System (MCS) is presented in this paper. The novelty of the approach consists in utilizing a signal of one sensor as the information source of a single member of the classifier ensemble. The size of the committee is delimited by the number of sensors applied for solving gas identification problems. The following base classifiers were considered: Support Vector Machine (SVM), the k-Nearest Neighbor (k-NN) method and two kinds of decision trees - CART and C4.5. Additionally, three fusion strategies were examined: majority voting, weight assignment based on the individual accuracy of the committee member and optimal weights combination found by the genetic algorithm. The MCSs performance was compared with the effectiveness of single classifiers which operated on the data set containing the response of the entire sensor array. The sensor signal compression by means of granulation was applied as the data pre-processing step. The classification problem consisted in recognizing volatile organic compounds (VOCs) in air, based on measurements performed by the array composed of fifteen semiconductor gas sensors. These devices were operated in the stop flow mode. Thus their signals were affected by many factors associated with altering exposure conditions, which enhanced the discrimination abilities of the sensors. (c) 2013 Elsevier B.V. All rights reserved.