VERIFAI
Verifiable and Explainable RIsk Forecasting Artificial Intelligence Framework
VERIFAI (Verifiable and Explainable RIsk Forecasting Artificial Intelligence) aims to verify the reliability of a risk assessment tool performing a binary classification forecast. The tool tries to predict whether or not a given event will occur in a future time window of a given length.
Hardware architecture: X64
Format: Dockerfile with Python and required libraries, Spin model checker and a bash entry point. along with all the source code available to run the experiment, and an artificial test subject. Make sure you have Docker installed, and run the built image with and an interface and terminal attached, using the -it flag.
More information: VERIFAI estimates whether the confidence level in the forecast made by the classifier, for a chosen particular instance, as opposed to averaged over the whole testing data set, is higher than a given user input safety threshold. The benefits of VERIFAI have been evaluated on a data set for the task of Cardio-Vascular Disease (CVD) risk assessment for emergency care patients, available on this plaform. The risk assessed is that of the patient returning to emergency care for CVD within 6 months following release from the hospital.
This is a joint contribution with the University Paris 1 Panthéon-Sorbonne.