“…a calibration relying on multiple sensor responses, gas multi-sensor devices have shown interesting capabilities for dealing with concentration estimation problems in complex mixtures. Often, the selected calibration model has been artificial neural networks [9][10][11], partial least squares (PLS) [12,13] and more recently Kernel methods like support vector machines [14,15]. Their capability to exploit partial selectivity of sensors as an advantage can give them a chance of well performing in modelling this complex framework.…”