External Cross Validation and Blinded Test
The Recognition Module construction and training process yields an estimate of the expected prediction correctness for
future classifications (i.e. an expected percentage of correct class assignments).
The "External Cross Validation" procedure provided by PATTERN EXPERT airspect
goes beyond that, by performing a complete simulation of "real-life" conditions.
Cyclicly, the software creates a complete Recognition Module with about 10% of your spectra excluded from all steps,
uses the finished module to classify the excluded spectra, and then checks the correctness of the individual class decisions. This cycle is repeated
10 times until, finally, all your spectra have been treated as unknown test data. As a result, you obtain a realistic (though quite conservative)
estimate of how the finally trained system will cope with actually unknown data in the future.
You may use this testing method instead of a true blinded test in cases where the total number of available samples is limited, because
in the External Cross Validation all the available spectra are utilized.
The results are presented in a graphical summary. For each of the ten testing blocks, you find the quality estimate yielded by the training process,
the absolute and the relative numbers of correctly resp. falsely assigned spectra of the non-trained testing set, as well as the total number
of features used for training and classifying. Moreover, you may check the system's behavior in the individual testing blocks in detail,
or you may compare the different features
(spectral locations) determined in the different blocks in a convenient comparison table.
Spectroscopy Techniques 
