The Kurvv index is a numeric value that ranges from 0 to 100 and represents the estimated accuracy of the predictions within the context of how the model output is intended to be used.
The formula to calculate the Kurvv index for each product or model may differ. Different methods used to calculate include ‘using well known validation metrics’ (i.e. F1 score), combining different known metrics all the way to ‘internally developed metric’. Final Kurvv index scores are scaled to range from 0 to 100.
Predictions with a Kurvv index of less then 50 is considered no better than a random guess, and is flagged and treated as such. 80 or above is considered accurate
It has been created for the purposed of (1) enabling non-technical users to quickly get a sense of how accuracy certain predictions are and (2) enabling a criteria through which predictions that does not meet performance requirements can be automatically filtered and flagged.
Its important to note that the Kurvv Index does not eliminate the need to calculate and consider the various validation metrics and scores used as best practices, since the Kurvv Index for each model has been developed based on assumptions derived from intended use & characteristics of the data source. For nuanced and a more comprehensive model evaluation based on sound data science practices please refer to best practices on the respective model.
Also, the numeric value are not intended to be used to calculate differences in accuracy between predictions. For example, a ‘80’ Kurvv index DOES NOT imply that it is twice as accurate as a prediction with ‘40’. Nor does ‘90’ imply that it is 12.5% more accurate than that of a ‘80’.