Personality Trait Recognition using ECG Spectrograms and Deep Learning.

Area of Science:

• Cardiology
• Psychology
• Artificial Intelligence

Background:

• Personality traits are complex and traditionally assessed via self-report questionnaires.
• Electrocardiogram (ECG) signals contain rich physiological information.
• Deep learning (DL) offers advanced methods for analyzing complex biological signals.

Purpose of the Study:

• To investigate the efficacy of ECG-derived spectrograms for personality trait recognition.
• To apply deep learning models for classifying the Big Five personality traits.
• To determine optimal parameters for ECG spectrogram generation.

Main Methods:

• Utilized the ASCERTAIN dataset with ECG recordings from 58 participants.
• Generated ECG-derived spectrograms with optimal window sizes.
• Employed Resnet-18 and Visual Transformer (ViT) for feature extraction and classification.
• Classified the Big Five personality traits: extraversion, neuroticism, agreeableness, conscientiousness, and openness.

Main Results:

• Achieved F1-scores consistently exceeding 0.9 for personality trait classification.
• Demonstrated the effectiveness of ECG spectrograms as informative features.
• Resnet-18 showed particular strength in discerning individual personality traits.

Conclusions:

• ECG signal spectrograms are a viable and informative modality for personality trait recognition.
• Deep learning models, particularly Resnet-18, can effectively classify personality traits from ECG data.
• This research opens new avenues for objective personality assessment using physiological signals.