Table 4.
List of sensors and their respective domains and metrics.
| Sensor | Metrics | Sense—Domain | Results | Reference |
|---|---|---|---|---|
| OpenFace | Facial Action Coding System (FACS) | Facial expressions analysis | State-of-the-art results for facial action unit recognition. Provides machine learning models for AU presence and intensity. | [54,55] |
| Facial Landmarks and EM | Chaos parameters, k-Nearest Neighbors (KNN) algorithm | Emotion estimation models | Strong correlation of chaos parameters with happiness. High accuracy (0.89) and ROC-AUC score (0.88) for emotion recognition. | [56] |
| AM-FED Dataset | Machine learning, chaos as a biomarker | Happiness estimation | Confirmed EM chaos as a biomarker for happiness, crucial even when the lower face is covered. | [57] |
| Face Mobility Index (FMI) | Face tracking, kNN | Facial impairment in PD | Statistically significant differences in facial impairment between healthy individuals and PD patients. AUC values between 88.9 and 88.4, F1 scores between 70.1 and 73. | [58] |
| Video Clips | Statistical shape model | Day-to-day variations in PD symptoms | Highlighted hypomimia in PD patients through decreased movement in expressions of happiness, disgust, and anger. | [59] |
| Computational Analysis | Movement Disorder Society’s Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) | Emotional facial expressions in PD | Reviewed computational techniques for measuring emotional facial expressions, with a deep learning model achieving 85% accuracy in masked face detection. | [60] |
| Iowa Gambling Task and Ekman 60 Faces Test | Voxel-based morphometry (VBM) | Neuropsychological deficits in PD | Correlation of OFC and amygdala degeneration with neuropsychological deficits in PD patients. | [61] |
| Adversarial Autoencoder and CNN-BiLSTM | Emotion recognition in speech and graphic representations | Emotion recognition | Achieved an accuracy of 0.99 in emotion classifications, demonstrating the effectiveness of advanced machine learning techniques. | [62] |
| EEG Features | SVM with automatic feature selection | Cross-subject emotion recognition | Mean recognition accuracy of 0.83 (AUC = 0.9), highlighting the potential of EEG features in emotional state biomarkers. | [63] |
| Text Analysis (DLSTA) | Deep Learning-Assisted Semantic Text Analysis | Emotion detection from text | Mean accuracy for emotion prediction of 0.83, with detection accuracy up to 0.92 and mean recall of 0.85. | [64] |