System and method for handiling a continuous attribute in decision trees
AssigneeMicrosoft Corporation
InventorsJeffrey R. Bernhardt, Pyungchul Kim, C. James MacLennan, David Maxwell Chickering
Life of this patent
Estimating the term…
Abstract
Continuous attributes are used as input attributes in decision tree creation. Buckets are created by dividing the range of values for the continuous attribute into sub-ranges of equal extent. These buckets form initial partitions. Mergers of adjacent partitions are considered to determine score gains from such mergers, and the most useful mergers occur. The resulting partitions are used as the discretization of the continuous attribute for use as an input attribute.
Claims48
A method for using a continuous attribute as an input attribute for a decision tree for a given output attribute, where a score can be calculated which corresponds to a measure of the usefulness of a subrange of an input attribute considered as a state over said given output attribute, comprising:
- determining a range of values for said continuous attribute;
- dividing said range into partitions, where each partition corresponds to a sub-range of said range;
- and
- determining whether to merge said at least two adjacent partitions.
A computer-readable medium comprising computer executable modules having computer executable instructions for using a continuous attribute as an input attribute for a decision tree for a given output attribute, where a score can be calculated which corresponds to a measure of the usefulness of a subrange of an input attribute considered as a state over said given output attribute, said modules comprising:
- a module for determining a range of values for said continuous attribute;
- a module for dividing said range into partitions, where each partition corresponds to a sub-range of said range;
- and
- a module for determining whether to merge said at least two adjacent partitions.
A computer device for using a continuous attribute as an input attribute for a decision tree for a given output attribute, Where a score can be calculated which corresponds to a measure of the usefulness of a subrange of an input attribute considered as a state over said given output attribute, comprising:
- means for determining a range of values for said continuous attribute;
- means for dividing said range into partitions, where each partition corresponds to a sub-range of said range;
- and
- means for determining whether to merge said at least two adjacent partitions.
Dependencies are inferred from the claim text.