Performance on prediction of concentration of gas C under drift-free conditions.

Two methods, linear PLS and non-linear SVR, were tested on the regression task of analyte C given at concentration 0.1, 0.4, 1 and 2 vol.%. Three arrays composed of 24 sensors, different in the types of sensor, were compared in terms of the root-mean-square error in prediction (RMSEP). For each array, the non-linear models outperform the linear models. All three arrays show similar performance with the SVR method, and it is hard to pick the best array.