Singh 2026 — FEEL: Quantifying Heterogeneity in Physiological Signals for Generalizable Emotion Recognition
Citation: Singh, P., Gupta, A., Jalan, S., et al. (2026). arXiv:2604.05926.
DOI: 10.48550/arxiv.2604.05926
Type: Benchmark paper (arXiv preprint)
Key Contribution
First large-scale multi-dataset benchmark for EDA + PPG emotion recognition. Unified evaluation of 16 architectures across 19 datasets. Provides the FEEL framework for standardized benchmarking.
Key Findings
| Result | Value |
|---|---|
| Best architecture | CLSP (contrastive signal-language pretraining) |
| Simple models competitive | RF, LDA, MLP in 36/114 top scores |
| Features vs raw | Handcrafted features consistently beat raw signals |
| Real-life → lab transfer | F1 = 0.79 |
| Lab → E4 transfer | F1 = 0.73 |
Relevance
Pre-built evaluation pipeline. Can directly run Meltdown Minder feature sets through the FEEL framework for standardized benchmarking.
Linked To
- Dataset: feel-benchmark
- Repo: feel-benchmark