Skip to main content
All CollectionsFAQData Management FAQ
How are variable predictors computed?
How are variable predictors computed?

Discusses the methodology used to identify which variables are most predictive

Updated over a week ago

During the Explore stage the G2M platform will give you the option to look at which variables in your dataset are most likely to be related to a variable of your choice. To do so, you will first need to select a variable in the variables list during the "Explore Variables" step.

The G2M platform will then scan all other variables in your variables list and compute the predictive power score (PPS) for each variable. The PPS score is an asymmetric, data-type-agnostic score that can detect linear or non-linear relationships between two variables. The score ranges from 0 (no predictive power) to 1 (perfect predictive power). Depending on the data type the PPS algorithm will either use regression or classification to score the relationship between two variables. The score assigned to the relationship will be a normalized mean absolute error (MAE) in the regression case, or a weighted F1 score in the classification case.

The PPS approach was developed by F. Wetschoreck, T. Krabel, and S. Krishnamurthy at 8080labs.

Did this answer your question?