Statistical assessment of quantification methods used in gas sensor system

This paper presents comparative study of neural, fuzzy, fuzzy neural and support vector machines approach to gas sensor array responses processing. Case study is quantification of toluene concentration in the environment with variable humidity. Sensor array containing TGS 800 and TGS 2000 series devices was exposed to gas mixtures to provide data sets for training, testing and validation of algorithms. Series of algorithms were prepared and tested using software written in MATLAB environment (partially in-house developed). The algorithms performance was compared and classic statistical test by McNemar was used to check whether the differences were significant or not. It was shown that response to this question may depend on the required quality of the sensor system. The results of statistical tests may be used in several ways to support the choice of the most suitable data processing method.