Explore one emotional profile
Set sex and emotion intensities, then request a prediction from the trained MLP.
Untouched holdout metrics
Performance measured on 85 real holdout cases not used for model training.
Confusion matrix
| Actual / Predicted | Negative | Positive |
|---|---|---|
| Negative | 32 | 1 |
| Positive | 7 | 45 |
Metrics describe predictive performance under the validation protocol and do not guarantee outcomes for individual cases.
How this app works
The PWA sends the selected profile to the public Predictor Lab API. The model, scaler, and database are not stored inside this app.
Inputs
Eight emotional channels plus sex are reordered internally to match the original training feature order before inference.
Model
Multilayer Perceptron (MLP) with the same trained weights and scaler used by the research predictor.
Privacy architecture
This frontend contains no private training artifacts and no Hugging Face access token. It calls a public prediction endpoint while the model artifacts remain in a private repository.
Connectivity
The app shell can be installed on a phone, but predictions require an internet connection because inference runs in Hugging Face.