Feature importance on classifier

[pecade]

"Variable importance" is a measure often used in classification models to identify which features (variables) are most useful in predicting the target variable. In general, this is done by fitting a model and then examining the model coefficients, or by using specific techniques, such as Permutation Feature Importance (PFI) or decision tree analysis.

dedupe trains a classifier model to distinguish between pairs of records that are duplicates and those that are not, but does not provide an easy method for directly examining the importance of variables. Instead, dedupe focuses more on providing a simple interface to perform the de-duplication task, and hides much of the internal details of the model. However, I think this is very valuable to get insihts on the variables definition to have a better understanding on feature interactions and further steps.