**6. Combination of gas sensors and pattern recognition methods**

According to an earlier review [6], receptors in mammalian olfactory systems do not show highly selective responses against specific analytes. Pattern recognition methods are thought to be a dominant mode used in processing signals from the broad responses of the mammalian olfactory system [6].

Cross-reactive chemical sensor arrays combined with pattern recognition methods to mimic mammalian olfactory systems have been studied as an alternative sensor system to traditional sensing devices that use a "lock-and-key" design [6]. In intelligent sensor arrays using pattern recognition methods, complex patterns generated by nonspecific cross-reactive sensors are analyzed for classification and identification of analytes [3, 6–8]. Cross-reactive sensor arrays are constructed using sensors that are responsive to a broad range of analytes and have differential sensitivity [3, 6]. Conventional semiconductor processes can be applied to miniaturize FET-based sensors for the fabrication of cross-reactive sensor arrays.

Before data analysis, complex signals obtained from sensor arrays can be preprocessed and normalized for the application of appropriate computational methods [7, 8, 10]. After preprocessing and feature extraction, the selected method is performed. Currently, there is no general rule for the selection of computational methods. Therefore, computational methods must be appropriately selected on the nature of the data and the particular situation [7].

Various types of gas sensor arrays have been used with pattern recognition methods [6–8], including FET-based gas sensor arrays. For example, Lundström et al. reported combination of catalytic-gate FET-based gas sensor arrays with pattern recognition methods [68, 69]. The signals from the FET-based sensor arrays were processed using conventional partial leastsquares regression and an artificial neural network to predict the concentrations of individual gases [69]. Molecularly modified Si NW-based FET sensors have also been combined with an artificial neural network to recognize VOCs and estimate their concentrations [70].
