Abstract
Oronasal thermal sensor signal recording within sleeping period was analyzed in sense of detection of respiratory events. The goal was to evaluate what extent an up to date nonlinear dimension reduction technique increases performance of apnea classification compared to classifications based on linear dimension reduction and raw data.
Based on an extensive database of recordings in apneic patients, we concluded that a non-linear approach didn't lead to better classification performance in comparison with a linear decomposition technique. We suggested, that it was due to the signal amplitude is naturally the main feature of respiratory events.
However, nonlinear methods do not naturally maintain the amplitude based structure in data. Due to this fact we get the worse signal representation in comparison with linear methods.