The Parafuel analyzer equipped with a fiber optic cable and probe was used to scan the samples at room temperature. The scanning wavelength range was 1200 nm to 2400 nm. A sample spectrum contained an average of 30 scans.
To build the calibration model, the spectra of the 37 samples were obtained in non-sequential order.  A spectrum of the probe in air provided the background from which absorbance processing was derived.   The probe allowed for sampling of the product directly in the collection jars. The ratio of Biodiesel ranged from 1 to 100% and included Animal Fat, Chicken Fat, Vegetable Fat, and Used Recycled Grease as FAME sources.  This can be seen in the wide variation in active areas of the first derivative spectra shown on the left below.  This extreme range of variables indicates the robustness of the technique.  Models on selected sources and/or specific ranges of interest would provide even better performance.
The performance statistics of the model were derived from 'leave one out' cross validation.  This technique removes each sample from the model in turn, recreating the model each time and predicting the removed sample as a true blind validation sample.  This technique is considered the most stringent of validation methods. In addition to measuring the ratio of biodiesel, this sample set afforded the opportunity to model the identification or discrimination of the FAME source.  Identification of Animal, Chicken, Vegetable, and Recycled oils was 100% accurate.