| CPC G06F 21/6245 (2013.01) | 20 Claims |
|
1. A computer-implemented method comprising:
analyzing a predictive model and input data for the predictive model using an explainability algorithm, the analyzing resulting in at least a feature importance value of a feature and an associated priority ranking, wherein the associated priority ranking is adjusted responsive to an assigned weight based at least in part on a sensitivity value of the feature;
analyzing feature values of the feature using a generalization function, wherein the analyzing of the feature values results in generation of a set of candidate feature values;
determining, based on the set of candidate feature values and on the associated priority ranking, an alternative feature, wherein the alternative feature is a generalization of the feature, wherein the alternative feature is a generalized feature having a generalized domain, wherein each feature value in the generalized domain corresponds to one or more feature values in a domain of the feature, wherein a number of feature values in the domain is greater than a number of feature values in the generalized domain, whereby the generalized feature is a generalization of the feature;
comparing an accuracy of the predictive model to a threshold performance value, wherein the accuracy is based on outputs of the predictive model using the alternative feature; and
mapping, responsive to the accuracy being above the threshold performance value, feature values in the input data that are in the set of candidate feature values to a generalized representative value in the generalized domain.
|