How Predictive Model Accuracy is Measured

You can view latest accuracy information for the voluntary-termination and performance predictive models on the Manage Predictive Models page.

Predictive Model Accuracy

The predictive models for both voluntary termination and performance are built using a subset (approximately 70%) of the available historical data. Oracle Data Mining (ODM) tests the accuracy of the models by making voluntary-termination and performance predictions for the remaining held-aside data (the 30% not used in building the predictive models). ODM then compares its predictions with actual outcomes.

The percentage accuracy of the predictive model:

• For voluntary-termination predictions, derives from the percentage of correct predictions made for all employees, both those who leave the enterprise and those who remain.

• For performance predictions, is a measure of how closely the predicted values match the actual values.