Key result
Logistic regression with GSR and PPG achieves ~84% emotion recognition accuracy, outperforming Random Forest and KNN.
Observational (n=30)
Combining mobile physiological sensing with computational intelligence provides a practical and non-invasive approach to emotional health monitoring.
Answers are grounded in this paper and cite their sources.
May facilitate non-invasive emotion monitoring; leaves open validation before clinical or research adoption.
This study presents a portable multimodal system for emotion recognition during Quranic recitation using Galvanic Skin Response (GSR) and mobile Photoplethysmography (PPG) via the Welltory application. Thirty healthy Arabic-speaking participants listened to selected Quranic verses while physiological signals were recorded. A validated data augmentation strategy (p > 0.05) expanded the initial dataset to 480 samples from 12 participants. Six machine learning models were evaluated using Leave-One-Subject-Out (LOSO) cross-validation. Logistic Regression achieved the highest performance, with accuracies of 83.9% for arousal and 82.3% for valence, significantly outperforming Random Forest and KNN (p < 0.05). Particle Swarm Optimization (PSO) determined optimal emotional thresholds, improving classification over fixed-threshold methods (p $$=$$ 0.018). GSR proved effective for detecting rapid arousal changes, while PPG offered broader HRV-derived insights and greater mobile usability. A PPG-only model successfully predicted both arousal and valence, enabling future smartphone-only deployment. Emotional responses on the arousal-valence circumplex clustered into three categories: Joy (33.3%), Anger (33.5%), and Pleasure (33.1%), corresponding to gamma-band frequencies in the audio stimulus. These findings demonstrate that combining mobile physiological sensing with computational intelligence provides a practical, culturally relevant, and non-invasive approach to emotional health monitoring.
No takes yet. Share an insight, caveat, or question.
Darabkh et al. (2026) conducted an observational in Healthy (n=30). Machine learning emotion recognition using GSR and PPG vs. Other machine learning models (Random Forest, KNN) and fixed-threshold methods was evaluated on Emotion recognition accuracy for arousal and valence (p=<0.05). Logistic Regression using GSR and PPG signals achieved emotion recognition accuracies of 83.9% for arousal and 82.3% for valence, significantly outperforming Random Forest and KNN (p < 0.05).
Synapse has enriched 10 related papers. Consider them for comparative context: