Predictive Modeling
in Healthcare
We apply our AI simulation process to the medical domain, where we make predictions that can save lives and money.
Crucial outcomes are often relatively rare events in the sea of healthcare data.
We use Generative Adversarial Networks (GANs) to create synthetic examples of rare events so that our machine learning algorithms can identify who is at risk and what mitigations are likely to be most helpful.
For example, we are using medical claims data to identify which patients are likely to overdose or become addicted to opioids
We have this idea of a machine learning model ecosystem. Predictive models are not sufficient; you need models that help supply them with data and put their predictions in context.
- Generative model creates synthetic data records to amplify rare events
- Predictive model predicts the probability of events
- Explanatory model predicts how the predictive model will make decisions around data points to generate explanations
- Certainty model predicts the probability that the predictive model is correct
