EVA-MED: An Enhanced Valence-Arousal Multimodal Emotion Dataset for Emotion Recognition

TL;DR

EVA-MED addresses the challenge of precise $Valence$-$Arousal$ modeling while incorporating individual differences by collecting multimodal physiological data (EEG, ECG, PI) from 64 participants under two emotion elicitation paradigms: video-based valence induction and the Mannheim Multicomponent Stress Test ($MMST$). The dataset combines controlled laboratory measurements with extensive psychosocial assessments and provides a full preprocessing and validation pipeline, enabling robust personalized emotion recognition research. A 1D-CNN trained on EEG features achieves top-tier performance (Valence $=90.46\%$, Arousal $=93.44\%$), with ECG and PI providing complementary, albeit lower, accuracy ($ ext{Valence}_{ECG}=82.00\%$, $ ext{Arousal}_{ECG}=86.65\%$; $ ext{Valence}_{PI}=73.23\%$, $ ext{Arousal}_{PI}=73.92\%$). This resource supports advances in personalized affective computing and real-world emotion-aware systems, including long-term wearable monitoring for mental health and human-computer interaction applications, by enabling precise, individual-aware modeling of emotional dynamics across contexts.

Abstract

We introduce a novel multimodal emotion recognition dataset that enhances the precision of Valence-Arousal Model while accounting for individual differences. This dataset includes electroencephalography (EEG), electrocardiography (ECG), and pulse interval (PI) from 64 participants. Data collection employed two emotion induction paradigms: video stimuli that targeted different valence levels (positive, neutral, and negative) and the Mannheim Multicomponent Stress Test (MMST), which induced high arousal through cognitive, emotional, and social stressors. To enrich the dataset, participants' personality traits, anxiety, depression, and emotional states were assessed using validated questionnaires. By capturing a broad spectrum of affective responses while accounting for individual differences, this dataset provides a robust resource for precise emotion modeling. The integration of multimodal physiological data with psychological assessments lays a strong foundation for personalized emotion recognition. We anticipate this resource will support the development of more accurate, adaptive, and individualized emotion recognition systems across diverse applications.

Figures (6)

• Figure 1: Experiment Procedures.
• Figure 2: SAM Diagram.
• Figure 3: Experimental Setup for Multimodal Physiological Data Collection. Participants wore an EEG cap to record brain activity, while ECG electrodes were placed on the chest and lower rib area to monitor heart activity. A wrist-worn wearable device was used to collect pulse interval (PI) data. The participant performed tasks on a computer while physiological signals were continuously recorded.
• Figure 4: Pulse Interval.
• Figure 5: Dataset Structure.