System and method for feature selection in decision trees

A method for selecting output attributes for use in a decision tree from a set of possible output attributes and for selecting input attributes for use in said decision tree from a set of possible input attributes comprising:

• determining an interestingness score for each output attribute based on the difference between the entropy of the output attribute E(A) and a most favored entropy value m;
• selecting at least one output attribute for use in said decision tree from said set of possible output attributes based on said interestingness scores;
• determining a score gain sum for each input attribute based on the sum of gain scores over all of said selected at least one output attribute for use in said decision tree;
• and
• selecting at least one input attribute for use in said decision tree from said set of possible input attributes based on said score gain sums.

A computer-readable medium comprising computer-executable modules having computer-executable instructions for selecting output attributes for use in a decision tree from a set of possible output attributes and for selecting input attributes for use in said decision tree from a set of possible input attributes, said modules comprising:

• a module for determining an interestingness score for each output attribute based on the difference between the entropy of the output attribute E(A) and a most favored entropy value m;
• a module for selecting at least one output attribute for use in said decision tree from said set of possible output attributes based on said interestingness scores;
• a module for determining a score gain sum for each input attribute based on the sum of gain scores over all of said selected at least one output attribute for use in said decision tree;
• and
• a module for selecting at least one input attribute for use in said decision tree from said set of possible input attributes based on said score gain sums.

A computer device for selecting output attributes for use in a decision tree from a set of possible output attributes and for selecting input attributes for use in said decision tree from a set of possible input attributes, comprising:

• means for determining an interestingness score for each output attribute based on the difference between the entropy of the output attribute E(A) and a most favored entropy value m;
• means for selecting at least one output attribute for use in said decision tree from said set of possible output attributes based on said interestingness scores;
• means for determining a score gain sum for each input attribute based on the sum of gain scores over all of said selected at least one output attribute for use in said decision tree;
• and
• means for selecting at least one input attribute for use in said decision tree from said set of possible input attributes based on said score gain sums.