classification_helpers: Classification - General Considerations
In bpollner/aquap2: Multivariate Data Analysis Tools for R including Aquaphotomics Methods

classification_helpers

R Documentation

Classification - General Considerations

General remarks regarding the use of the classification functions in package aquap2.

Except the svm classification, most methods do not work well in fat data matrices, so if there are more variables (wavelengths) than observations. For an easy reduction of the number of wavelengths we provide, especially in the Aquaphotomics context, a special function to reduce the number of wavelengths in a dataset (as produced by gfd): function siWlg can isolate any custom group of wavelengths from the dataset, or, if left at its default, is isolating the ranges of the 12 water matrix coordinates within the 1st overtone. For further data reduction, this groups can be averaged within as well by setting the second argument in siWlg to TRUE.
For reducing the number of variables not in the raw data, but on the classification side of the algorithm, it is possible to apply all classifiers not on the rawdata, but on the PCA scores of the rawdata. This option can be activated by setting the respective argument .pcaRed in the analysis procedure to TRUE. In this case the prediction data for crossvalidation as well as the independent test data are projected into the pca-models of the training data, and the resulting scores are then used for classification.
For both traditional crossvalidation and the bootstrap "crossvalidation", consecutive scans of the same sample are always excluded resp. included together.
A traditional crossvalidation (no bootstrap) is indicated via a "." after the name of the classificaiton-variable, while the bootstrap process is indicated via a "'".