4.1. Introduction

4.1.1. Introduction

Detecting stress is crucial across fields such as healthcare, adaptive technology, and mental well-being, where timely intervention can improve outcomes and enhance user experiences. This chapter presents our approach to classifying stress using EEG and ECG signals from the DREAMER dataset.

We start with a brief overview of the problem, emphasizing the unique value of physiological data for stress detection. EEG and ECG signals together can provide a detailed picture of the body’s response to stress—a response often beyond conscious control and therefore highly reliable. The DREAMER dataset, chosen as the foundation for this work, provides synchronized EEG and ECG recordings, along with participant-reported emotional ratings, will be used as the foundation for this study.

Next, we outline the data processing and system design pipeline, which includes data preparation, preprocessing, and feature extraction steps. Following this, we cover model selection, training, and evaluation, identify the suitable approach for distinguishing stress from non-stress states. Finally, we review classification performance, highlight challenges, and suggest future enhancements.

4.1.2. Problem Statement

Stress is a on-going critical issue affecting mental and physical well-being, with long-term impacts on health , . Existing methods for stress detection often rely on subjective self-reports or resource-intensive clinical devices, making stress detection challenging in everyday settings. This project seeks to develop an automated approach for stress detection using EEG and ECG signals. By focusing on brain-heart dynamics, this study aims to classify emotional states as either stress or non-stress, using features from EEG and ECG data to create a more accessible, objective tool for stress assessment.

4.1.3. Background of DREAMER Dataset

The DREAMER dataset is a multi-modal database developed for emotion recognition, capturing EEG and ECG signals recorded from 23 participants during exposure to 18 audiovisual stimuli. Each stimulus aimed to evoke specific emotions, allowing participants to self-rate their emotional states across dimensions of valence, arousal, and dominance. The dataset’s use of low-cost, wireless equipments makes it suitable for research that aims to detect emotional states in non-clinical environments. This characteristic aligns well with the objectives of this project, which focuses on understanding the physiological underpinnings of stress and developing a classification model based on these signals.

4.1.4. Scope of Stress Classification

This study aims to detect stress by selecting specific emotions from the nine available targets in the DREAMER dataset, constructing a binary classification of stress versus non-stress. Emotions such as anger and fear, which are known to provoke stress responses, are grouped under the "stress" label. In contrast, calmness is chosen to represent non-stress and is labeled accordingly. By narrowing the focus to these key emotional states, we aim to learn distinct patterns in EEG and ECG responses associated with stress.