Reasoning classification based on feature pertubation

A system for identifying informative features of a data set, the system comprising:

• at least one processor and at least one memory device;
• a perturber configured to modify the data set by removing all but one feature of a plurality of features in the data set and to replace the one feature in the data set with a different feature to create at least one perturbed data set;
• a classifier for a machine learning system configured to classify the data set according to a set of rules, the classifier further configured to classify the at least one perturbed data set according to the set of rules;
• and
• a feature comparison module configured to receive a first result from the classifier for the data set and a second result from the classifier for the at least one perturbed data set and compare the first result with the second result to generate a representation of a difference between the first and the second result.

A computer readable storage device excluding transitory media and having computer executable instructions that when executed cause at least one computing device to:

• train a classifier for a machine learning system to classify data according to a set of rules;
• process a first data set through the classifier to obtain a first classification for the first data set;
• identify a second data set;
• perturb the first data set by changing or removing a single feature of a plurality of features in the first data set, and perturb the first data set with data from the second data set to obtain at least one perturbed data set;
• process the at least one perturbed data set through the classifier to obtain at least a second classification;
• compare the first classification with the second classification to determine a difference between the first classification and the second classification;
• and
• generate a representation of the difference between the first classification and the second classification.