MVPA performance significantly improved from using a large number of short runs, compared to several long runs, in across-run classifications with identical amounts of data.
After running classification algorithms, we tested the cross-validation error estimation using 10-fold cross-validation with 10 iterations.
It is therefore good practice to run multiple classifications specifying different numbers of output classes, e.g. 20, 50, and 100 classes.
Useful features in the run of classification accuracies with spectral resolution have been noted and discussed.
A value of j=201 was chosen to run the classification models.
For those techniques that were run in classification and regression mode, the respective class probability estimates (i.e. (hat{p}) and (bar{p})) ranked top.
These training signatures were distributed in 15 classes, which were merged later into 10 classes to run the classification.
Footfall pattern will be assessed throughout each 6 min run and classification will be based on the mode (ie, whichever pattern is most frequent) during the final 60 s.
To this end, we stepwise carried out cross-validated (leave-one-run-out) classifications and removed each time the 4% voxels with the lowest t-values obtained in the sound-localizer.
While the original problem of forming a ranking of the most likely classes can be solved by running several classification methods, the analysis presented here is moved one step further.
Two filter steps were applied before running the classification algorithms.
