Figure 1

Block diagram of the nested LOOCV procedure. In the inner most loop, the search for the best feature subset (for the 46 subjects in that loop) is performed by sequentially adding/removing features into/from the model, followed by an F-test to check the improvement made to the model. The middle loop is repeated 47 times, where each repetition returns a “final” model, as determined by HIST1 and HIST2, to the outer most loop. In the outer most loop, an LOOCV is performed where a test subject data was withheld and used to test the model returned by the middle loop, which has no prior information on the withheld data. The generalized performance was estimated from the 48 test results, and pooled histograms of HIST 1 and HIST2 were plotted to determine which features were most useful in estimating CO and SVR.