Predictive model for AD
AlzAware uses a pre-trained machine-learning model to calculate a cognitive risk reference score from structured health and demographic information.
Model built with
- More than 150,000 participants
- 24 predictors
- Cutting-edge AI approach: XGBoost classifier
Run a sample assessment result
This demo submits a pre-loaded survey profile to the prediction backend and displays the same reference-style result used in the public assessment page.
Model performance
- Randomly split data as 80/20 for training/testing
- Train a model using training data only
- Apply the trained model to the testing data for validation
- ROC-AUC values
0.872
Training AUC
0.861
Testing AUC
Output interpretation
- All participants' risk scores are predicted using the final model, and the score distributions for the two study samples, AD and non-AD, are visualized separately.
- The larger the score, the higher the model-estimated risk.
- Your predicted risk score is compared with the two reference samples to support decisions about whether professional cognitive evaluation may be worth considering.
