4.2. Dataset Overview

4.2.1. DREAMER Data Structure

The dataset is stored hierarchically in the DREAMER.mat file, which organizes data and metadata as follows:

Table 4-1: Hierarchical Structure of DREAMER Dataset in DREAMER.mat

Each participant's data includes EEG and ECG recordings during both the baseline (neutral state before each clip) and stimuli (emotion-inducing film clip) periods, along with self-reported ratings on arousal, valence, and dominance scales for each of the 18 film clips.

Signals were recorded using portable devices:

• Emotiv EPOC wireless EEG headset, offering sampling rates of 128 Hz for EEG

• Shimmer2 ECG sensor, offering sampling rates 256 Hz for ECG.

Table 4-2: Overview of DREAMER Dataset Characteristics

4.2.2. DREAMER Participant and Emotion Data

The DREAMER dataset includes emotional ratings from 23 participants, who evaluated each video clip on a five-point scale across valence (positivity), arousal (intensity), and dominance (sense of control). These three dimensions form the basis of the PAD (Pleasure-Arousal-Dominance) model, widely used in affective computing to map emotions in a 3D space.

The following Figure 1 and Figure 2 illustrate the PAD emotional state model in 3D space, along with the distribution of nine emotional states across valence, arousal, and dominance dimensions.

Figure 1: PAD Emotional State Model Representation

Figure 2: Distribution of Nine Emotional States in PAD Space

Each emotion’s placement within this space allows for classification into high/low categories or regression for specific scores. These ratings provides us a foundation for defining stress versus non-stress states in this study.