Hi BERGers!

This week (Wednesday 26th September), we have Guillermo Hidalgo Gadea giving a talk titled "Measuring Fatigue and Comfort: An Experimental Approach".

This talk will combine 2 exciting studies, which you can read a bit about in the abstracts below.

Looking forward to seeing you all there,

Gemma

This study aims to provide an interdisciplinary perspective on biosignal processing for driver monitoring systems. Machine learning models were trained with ECG, Pupil Diameter and Eyelid Opening data from overnight sleep deprived subjects experiencing microsleep in a supervised test track driving environment. Model performance of 0.85 accuracy and 0.96 precision indicates (a) promising results for noninvasive microsleep detection in real driving environment and (b) benefits of the presented interdisciplinary processing method for psychological research.

This paper presents the idea of brute force feature extraction for Electrocardiography (ECG) signals applied to discomfort detection. To build an ECG Discomfort Corpus an experimental discomfort induction was conducted. 50 subjects underwent a 2h (dis-)comfort condition in separate sessions in randomized order. ECG and subjective discomfort was recorded. 5min ECG segments were labeled with corresponding subjective discomfort ratings, and 6365 brute force features (65 low-level descriptors, rst and second order derivatives, and 47 function-als) and 11 traditional heart rate variability (HRV) parameters were extracted. Random Forest machine learning algorithm outperformed SVM and kNN approaches and achieved the best subject-dependent, 10-fold cross-validation results (r = .51). With this experiment, we are able to show that (a) brute force ECG feature sets achieved better discomfort detection than traditional HRV based ECG feature set; (b) cepstral and spectral ux based features appear to be the most promising to capture HRV phenomena.