To automatically discover an action attribute change plan by a rule from a database.
A method for changing a value of an action attribute to improve the desirability of a case is discovered by using a database including a state attribute being an attribute in which a decision maker is incapable of controlling a value, the action attribute being an attribute in which the decision maker is capable of controlling a value, and a result attribute; and a case evaluation function for evaluating the desirability of a case. An attribute space by the state attribute and the action attribute is divided so that a distribution of values of the result attribute satisfies a prescribed result attribute distribution reference in each region, and a plurality of state regions and a plurality of action regions are obtained. A plurality of policy evaluation state regions are generated by dividing a state space so that a value of the action attribute satisfies a prescribed action attribute distribution reference in each region. A plurality of detailed state regions are calculated by executing a region intersection calculation by using a plurality of the state regions and a plurality of the policy evaluation state regions. The method for changing a value of the action attribute to improve the desirability of an assembly of cases belonging to the detailed state region is calculated as a policy improvement rule being a mapping from the detailed state region to the action region.
Hidetoshi Tachibana
Yasukazu Sato
Hiroshi Yoshimoto
Yasushi Kawasaki
Jun Okazawa