Inteligentní metody pro rozpoznání a shlukování emocí na základě GSR a HRV signálů a systému fNRIS pro analýzu meditační fáze

Abstract

This master's thesis focuses on emotion recognition and meditation phase analysis using biosignals, specifically galvanic skin response (GSR), heart rate variability (HRV), RR intervals, functional near-infrared spectroscopy (fNIRS), and electroencephalography (EEG). The aim is to identify physiological changes related to meditation and classify emotional states based on multichannel biosignal recordings.

Healthy volunteers participated in a guided meditation protocol, during which biosignals were recorded and segmented into three phases: before meditation, during meditation, and after meditation. After preprocessing and filtering, time-series features were extracted using the tsfresh library and subsequently reduced using RFECV and LDA methods. The selected features were further analyzed using fuzzy clustering and statistical techniques.

The most informative features included statistical and spectral parameters. For GSR, commonly selected features included mean and quantile, reflecting electrodermal activity levels. In HRV, features such as RMSSD, as well as spectral components LF and HF, were frequently used. EEG and fNIRS contributed with features based on frequency band power and average concentrations of oxy- and deoxyhemoglobin in the prefrontal cortex.

Classification results confirmed that combining multiple biosignals (especially EEG, GSR, and fNIRS) improves accuracy in identifying meditation states compared to single-modality analysis. Fuzzy C-means clustering (FCM) was applied to group participants into two clusters based on differences between meditation and baseline phases. The results showed that some participants exhibited more pronounced and consistent responses, while others were less clearly assigned, reflecting individual variability in physiological reactions.

This work contributes to the development of intelligent systems for emotion analysis and shows potential for applications in biofeedback and mental health support.