Emotion detection unveiled: A cognitive-computational synthesis of physiological models, machine learning, and...

Area of Science:

• Affective computing
• Cognitive science
• Artificial intelligence

Background:

• Emotion recognition traditionally relied on basic machine learning, facing accuracy limitations (70-75%).
• A gap existed between physiological signal data and cognitive process understanding.
• Explainable AI (XAI) lacked a robust theoretical foundation in emotion recognition.

Purpose of the Study:

• To survey state-of-the-art advancements in physiological emotion recognition (2021-2025).
• To introduce a novel Cognitive-Computational Synthesis Framework (CCSF) linking physiological signals to cognitive processes.
• To provide a roadmap for developing robust, interpretable, real-time emotion recognition systems.

Main Methods:

• Comprehensive literature review of 40 key studies using PRISMA protocols.
• Analysis of the transition from traditional machine learning to deep learning architectures (Transformers, self-supervised, diffusion models).
• Evaluation of studies based on validation, generalizability, and adversarial robustness.

Main Results:

• Deep learning models now achieve >95% accuracy on datasets like SEED and DREAMER, surpassing older methods.
• The CCSF framework maps physiological signals (EEG, ECG, GSR) to cognitive functions (attention, arousal, bias).
• Significant improvements in cross-subject generalizability and adversarial robustness were observed.

Conclusions:

• The CCSF provides a theoretical basis for XAI in affective computing.
• Modern deep learning architectures represent a paradigm shift in physiological emotion recognition.
• The survey offers strategic insights for future research in robust and interpretable emotion AI.