PPG-Based Affect Recognition with Long-Range Deep Models: A Measurement-Driven Comparison of CNN, Transformer, and Mamba Architectures

2026-04-30 · Source: cs.LG updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Medical Devices & Health Technology · Depth: Expert, quick

Summary

A study compared four deep learning architectures—Convolutional Neural Networks (CNN), CNN-Long Short-Term Memory (LSTM) hybrid, Transformers, and Mamba—for classifying arousal, valence, and relaxation states from wrist-based Photoplethysmography (PPG) signals. The research evaluated these models using a subject-independent 5-fold cross-validation protocol, ensuring identical preprocessing, segmentation, and training pipelines. Results indicate that while Transformer and Mamba models achieved performance comparable to a CNN baseline, they did not consistently outperform it across all affect recognition tasks. CNNs emerged as the most effective overall, offering the highest accuracy with the smallest model size. Transformers demonstrated a better balance of F1 scores for Arousal and Relaxation, suggesting specific strengths. This work provides the first evaluation of Transformer and Mamba models for PPG-based affect recognition, offering practical guidance for wearable affective monitoring systems.

Key takeaway

For research scientists developing wearable affective monitoring systems, you should prioritize Convolutional Neural Networks (CNNs) for PPG-based affect recognition due to their superior accuracy and smaller model footprint. While Transformers and Mamba models offer comparable performance, CNNs provide a more efficient solution, especially given typical small and noisy PPG datasets. Consider Transformers if your application specifically requires a balanced F1 score across arousal and relaxation states.

Key insights

CNNs remain highly effective for PPG-based affect recognition, often outperforming newer long-range models.

Principles

• CNNs offer highest accuracy with smallest model size.
• Transformers balance F1 scores for Arousal and Relaxation.

Method

Four deep learning architectures were compared for PPG-based affect recognition using a subject-independent 5-fold cross-validation with identical pipelines.

In practice

• Consider CNNs for compact, accurate PPG affect recognition.
• Evaluate Transformers for balanced Arousal/Relaxation F1 scores.

Topics

• PPG-based Affect Recognition
• Deep Learning Architectures
• Convolutional Neural Networks
• Transformer Models
• Mamba State-Space Models

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, AI Engineer

Related on AIssential

• Emotional regulation improves deep learning-based image classification — Emotional Regulation enhances deep learning image classification by modeling emotion through artificial subjective experience.
• Transformers Step-by-Step Explained (Attention Is All You Need) — Transformers enable parallel processing and efficient context capture in sequential data through a dynamic attention mechanism.
• Deep Temporal Modeling and Ensemble Fusion for Multimodal Emotion Recognition from Physiological Signals — Combining deep temporal models with sensor and ensemble fusion significantly boosts physiological emotion recognition accuracy.
• CLIMB: Controllable Longitudinal Brain Image Generation using Mamba-based Latent Diffusion Model and Gaussian-aligned Autoencoder — CLIMB uses Mamba-based latent diffusion and a Gaussian-aligned autoencoder for efficient, controllable longitudinal brain MRI generation.
• Boundary-aware Prototype-driven Adversarial Alignment for Cross-Corpus EEG Emotion Recognition — The PAA framework improves cross-corpus EEG emotion recognition by aligning class-conditional distributions and refining decision boundaries.
• How to Fine-Tune an SLM for Emotion Recognition — Fine-tuning SLMs for fine-grained emotion recognition on imbalanced datasets requires a multi-technique approach to data balancing and loss weighting.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.LG updates on arXiv.org.