Selection of optimal sensor/temperature conditions for winegrape analysis using generalized additive modeling of thermally cycled metal oxide sensors

For the first time, generalized additive models (GAMs) were applied to the analysis of data from thermally cycled metal oxide sensors (TC-MOXs). These were used to evaluate Riesling grape juices of low, medium and high grades harvested from different blocks of a single parcel. The results obtained with GAMs were compared to a more empirical model (Fort´s model). GAMs fitted the sensor responses better than Fort´s model. Using GAMS, log-resistance modeled as a smooth function of time and sensor number correctly classified 75% of the samples. Gas chromatography-mass spectrometry analysis clearly discerned differences among the three grape qualities.