One can then expect that with OLS clustering the particularities of each regime is

better identified and modeled even if they are represented by a few points only, when

those few points might not have as much weight with k-means and be "blended" to a

behavior occurring statistically more often.


 1



 500 1000 1500 2000 2500 3000

 00


 500 1000 1500 2000 2500 3000



 i i -- I 
 500 1000 1500 2000 2500 3000
 1 5 -

 500 1000 1500 2000 2500 3000
 50

 50 I-----------I-----I-------I-------I--


 500 1000 1500 2000 2500 3000


Figure 4-12 Results: (a) Flow: desired in blue, predicted in red; (b) MSE criterion, full
 line is for expert 1, dotted line is for expert 2; (c) Winner; (d) Pressure.

 Figure 4-13 shows results for approach III, but since all mandatory breaths were

detected accurately in approach II, the spike detection does not change much the history

of segmentation, and the waveforms of the criterion are a lot similar to those in approach

II. Also the spike detection mechanism does not help for a sooner detection of the switch

back to regime 2, because expert 1 does not show such big spikes when a change occurs

as expert 2.