Synheart Emotion: Privacy-Preserving On-Device Emotion Recognition from Biosignals

2025-11-09

AI Summary

To address privacy leakage and high latency in cloud-based affective computing systems, this paper proposes an edge-native real-time emotion recognition method tailored for wrist-worn devices. Leveraging photoplethysmography (PPG) signals acquired from the wrist, the approach replaces deep learning models with a lightweight ExtraTrees ensemble learning framework and employs ONNX-based optimization for model compression and inference acceleration. Evaluated on the WESAD dataset, the unimodal solution achieves an F1-score of 0.623 for stress/emotion classification, with a model size of only 4.08 MB and an edge-side inference latency of merely 0.05 ms—152× faster than baseline methods and 30.5% more storage-efficient. This work is the first to empirically demonstrate the superiority of classical tree-based models for small-sample physiological-signal-based emotion recognition, establishing a practical, privacy-preserving, and resource-efficient technical pathway for wearable applications under stringent computational constraints.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Cognitive Modeling & Cognitive Systems: Affective Computing
• Computer Vision: Biometrics, Face, Gesture & Pose

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Human-computer interaction increasingly demands systems that recognize not only explicit user inputs but also implicit emotional states. While substantial progress has been made in affective computing, most emotion recognition systems rely on cloud-based inference, introducing privacy vulnerabilities and latency constraints unsuitable for real-time applications. This work presents a comprehensive evaluation of machine learning architectures for on-device emotion recognition from wrist-based photoplethysmography (PPG), systematically comparing different models spanning classical ensemble methods, deep neural networks, and transformers on the WESAD stress detection dataset. Results demonstrate that classical ensemble methods substantially outperform deep learning on small physiological datasets, with ExtraTrees achieving F1 = 0.826 on combined features and F1 = 0.623 on wrist-only features, compared to transformers achieving only F1 = 0.509-0.577. We deploy the wrist-only ExtraTrees model optimized via ONNX conversion, achieving a 4.08 MB footprint, 0.05 ms inference latency, and 152x speedup over the original implementation. Furthermore, ONNX optimization yields a 30.5% average storage reduction and 40.1x inference speedup, highlighting the feasibility of privacy-preserving on-device emotion recognition for real-world wearables.

Problem

• Developing privacy-preserving on-device emotion recognition from biosignals
• Evaluating machine learning models for wrist-based PPG emotion detection
• Optimizing models for real-time wearable applications with minimal latency

Innovation

• On-device emotion recognition from wrist PPG signals
• ExtraTrees ensemble method outperforms deep learning models
• ONNX optimization enables efficient deployment on wearables