Raspoznavanje stresa na temelju analize fizioloških značajki metodama dubokog učenja

undergraduate thesis

• Title (english): Stress recognition based on the analysis of physiological features using deep learning methods
• Author: Ljubica Nikolić
• Mentor: Siniša Popović (mentor)
• Committee member: Siniša Popović (predsjednik povjerenstva)
• Committee member: Hrvoje Nuić (član povjerenstva)
• Committee member: Mate Gambiraža (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-07-03, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Ovaj rad se bavi prepoznavanjem stresa temeljenog na analizi fizioloških značajki koristeći metode dubokog učenja. Korištena je javno dostupna baza podataka WESAD (Wearable Stress and Affect Detection), koja sadrži multivarijatne podatke prikupljene putem uređaja RespiBAN. To se odnosi na skup signala koji obuhvaćaju elektrokardiogram (ECG), elektrodermalnu aktivnost (EDA), elektromiogram (EMG), temperaturu (TEMP), akceleraciju u 3 smjera (ACC; 3 signala) i respiraciju (RESP). Podaci su prethodno obrađeni i normalizirani kako bi mogli poslužiti kao ulaz u modele auto-enkodera. Razvijeni su modeli koji koriste sirove podatke i modeli koji koriste valićno transformirane podatke. Rezultati su pokazali da model koji koristi sirove podatke ima bolju točnost od 79.3 % te F1 rezultat od 78.5 % u prepoznavanju stresa u usporedbi s modelom koji koristi valićnu transformaciju kojem je točnost 71.5 % te F1 rezultat 73.3 %.

Abstract (english)

This paper deals with stress recognition based on the analysis of physiological features using deep learning methods. The publicly available WESAD (Wearable Stress and Affect Detection) database was used, containing multivariate data collected via RespiBAN devices. This pertains to a set of signals encompassing electrocardiogram (ECG), electrodermal activity (EDA), electromyogram (EMG), temperature (TEMP), acceleration in 3 directions (ACC; 3 signals), and respiration (RESP). The data was preprocessed and normalized to serve as input to autoencoder models. Models utilizing raw data and models employing wavelet-transformed data were developed. The results indicated that the model using raw data achieved higher accuracy of 79.3 % and F1 score of 78.5 % in stress recognition compared to the model using wavelet transformation, which achieved an accuracy of 71.5 % and F1 score of 73.3 %.

• Keywords: detekcija stresa; duboko učenje; afektivno računarstvo; auto-enkoder; obrada multimodalnih signala; diskretna valićna transformacija; WESAD
• Keywords (english): stress detection; deep learning; affective computing; autoencoder; multivariate signal processing; discrete wavelet transform; WESAD
• Language: croatian
• URN:NBN: urn:nbn:hr:168:958526
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2024-10-12 21:40:59