Přístupnostní navigace
E-application
Search Search Close
Bachelor's Thesis
Author of thesis: Filip Sedlár
Acad. year: 2025/2026
Supervisor: Ing. Andrea Němcová, Ph.D.
Reviewer: Ing. Enikö Vargová
This bachelor’s thesis addresses automated stress detection using multimodal physiological data in an office environment. Such detection is deemed necessary given the health and economic impacts of mental stress. The theoretical framework examines the physiological mechanisms of the stress response and evaluates the biosignals, algorithms, and methodologies best suited for unobtrusive monitoring. In the practical part, a data processing pipeline was developed using the SWELL-KW dataset. Raw electrocardiogram (ECG) and electrodermal activity (EDA) signals were synchronised, segmented, and pre-processed using a custom artefact correction, convex optimisation-based EDA decomposition, and subject-specific Z-score normalisation. From this, a comprehensive set of time, frequency, and non-linear features was extracted. To combat high dimensionality, a non-parametric Kruskal-Wallis filter combined with Minimum Redundancy Maximum Relevance (mRMR) and Random Forest (RF) algorithms filtered an optimal subset of 3 key features. A granular ablation study mathematically validated an expansion towards 4 key features, proving that this dimensionality is optimal and that excessive features provide negligible predictive gains while introducing unnecessary computational load which would drain wearable devices. A RF ensemble utilising the mRMR feature subset emerged as the final deployable winning model due to its high computational efficiency and resistance to overfitting. Under a strict subject-independent protocol, it achieved an Accuracy of 82.93%, Sensitivity of 79.04%, and a Specificity of 88.25% on the unseen Test set. Ultimately, this thesis demonstrates that a compact, mathematically validated set of non-linear features enables highly reliable stress detection without requiring complex deep learning architectures.
Stress detection, multimodal biosignals, SWELL-KW, ECG, EDA, HRV, feature selection, machine learning, mRMR, random forest (RF), support vector machine (SVM), subject-independent validation
Date of defence
16.06.2026
Result of the defence
Defended (thesis was successfully defended)
Grading
A
Process of defence
Student prezentoval výsledky své práce a komise byla seznámena s posudky. Ing. Smital položil otázku, zda student pracoval s více databázemi, jakým způsobem byl v použité databázi vyvoláván stres a zda byla součástí experimentu také fyzická aktivita. Ing. Jakubíčková položila otázku, v jakém programovacím jazyce byla práce realizována. Student obhájil bakalářskou práci a odpověděl na otázky členů komise a oponenta.
Language of thesis
English
Faculty
Fakulta elektrotechniky a komunikačních technologií
Department
Department of Biomedical Engineering
Study programme
Biomedical Technology and Bioinformatics (BPC-BTB)
Composition of Committee
doc. Ing. Petr Kudrna, Ph.D. (předseda) Ing. Markéta Jakubíčková, Ph.D. (místopředseda) MUDr. Zuzana Nováková, Ph.D. (člen) Ing. Lukáš Smital, Ph.D. (člen) Ing. Vratislav Harabiš, Ph.D. (člen) Ing. Larisa Chmelíková, Ph.D. (člen)
Supervisor’s reportIng. Andrea Němcová, Ph.D.
Grade proposed by supervisor: A
Reviewer’s reportIng. Enikö Vargová
Grade proposed by reviewer: A
Responsibility: Mgr. et Mgr. Hana Odstrčilová