Assessing Emotions via Machine Learning and Affordable Wearables
The latest research has shed light on the effectiveness of the MUSE headband and Shimmer GSR+ devices in predicting emotional states during stimuli exposure, moving beyond controlled lab environments and into real-world applications.
These non-invasive devices, designed for a wide population, offer a promising avenue for Brain Computer Interfaces and Affective Computing to impact daily life.
The MUSE headband primarily records Electroencephalography (EEG) signals from frontal and temporal sites, making it ideal for detecting brainwave patterns associated with emotional processing. On the other hand, the Shimmer GSR+ device measures Galvanic Skin Response (electrodermal activity), reflecting sympathetic nervous system arousal, which is highly correlated with physiological arousal but less so with valence.
When it comes to predicting arousal and valence, the Shimmer GSR+ signals are generally more reliable for arousal prediction due to their strong correlation with sympathetic nervous system activation related to arousal levels. EEG data from the MUSE headband, however, is often more informative for valence, as frontal asymmetry and other EEG features have been linked to positive vs. negative affective states.
In typical performance with Machine Learning, accuracies depend on features and classifiers. For arousal, using GSR features alone, accuracies in the range of 70-85% are commonly reported for binary classification (high vs. low arousal). EEG-based valence classification with MUSE data typically yields accuracies around 65-80% in binary valence classification scenarios.
Combining EEG (MUSE) and GSR (Shimmer) features often improves performance due to complementary information, potentially achieving 75-90% classification accuracy depending on stimulus type, participant variability, and Machine Learning models.
Machine learning predictions usually show moderate to strong correlations with self-assessed valence and arousal, but rarely perfectly match since physiological signals reflect implicit states that can differ from conscious self-report. Accuracy or correlation coefficients (e.g., Pearson's r) between predicted labels and self-report ratings often fall in the 0.5 to 0.75 range, indicating reasonably good but not perfect agreement.
In summary, the MUSE headband and Shimmer GSR+ devices show promising potential in predicting emotional states, particularly when combining EEG and GSR features, improving accuracy and correlation with self-assessment.
| Prediction Target | Device/Data | Typical Accuracy | Relation to Self-Assessment | |-------------------|-------------|------------------|----------------------------------------| | Arousal | Shimmer GSR+ | ~70-85% | Moderate to strong correlation (~0.5-0.7) with self-report | | Valence | MUSE EEG | ~65-80% | Moderate correlation with subjective valence ratings | | Combined | EEG + GSR | ~75-90% | Improved accuracy, better correlation with self-assessment|
It's important to note that performance varies greatly with stimulus type, participant number, and Machine Learning approach. More complex models, like deep learning, sometimes outperform classical Machine Learning but require more data. Some studies also show that physiological measures capture instantaneous affective changes, which self-reports may miss.
If you're working on a similar project, I can suggest appropriate Machine Learning models, feature extraction methods, and evaluation protocols to optimize prediction accuracy with these devices.
Science and health-and-wellness are intertwined in the use of the MUSE headband and Shimmer GSR+ devices for mental health monitoring. These non-invasive tools, powered by technology, have shown promising potential in predicting emotional states, helping bridge the gap between lab environments and real-world applications. Moreover, the combined use of EEG (MUSE) and GSR (Shimmer) features can significantly improve accuracy and correlation with self-reported emotional states, offering innovative opportunities in improving overall mental health and wellness.
