Data analysis for a hybrid sensor array

We present the results obtained in measuring diverse food products with the Moses II electronic nose (EN) equipped with three different classes of chemical sensors; namely, seven quartz micro-balances (QMB), eight semiconductor sensors (S) and four electrochemical cells (EC). Data are analyzed both with traditional PCA plots, pointing out the limits encountered by this technique and via exhaustive sensor selection. The principal sensor selection results are that: (a) the ranking of the sensor type with regard to discrimination is QCM > EC > S; (b) selected hybrid sensors have much better performances than selected sensors from any single sensor class (test set error lowered by circa 35%); (c) sensors selected from the hybrid array also have better performances than the complete set of hybrid sensors (test set error lowered by circa 25%); and in particular (d) a subsets of as few as two sensors (one QCM, one EC cell) give results similar or better to all 19 sensors.