| 摘要 |
<p><P>PROBLEM TO BE SOLVED: To create a decision tree by taking calculation costs as well as classification accuracy into account to shorten processing time for data classification. <P>SOLUTION: A feature amount calculation unit 2 calculates a feature amount from learning data to calculate mean costs (calculation costs) required for calculating the feature amount. A decision tree learning unit 3 calculates an evaluation value on the basis of an impurity degree indicating lowness of the classification accuracy to the learning data and an inverse number of the calculation costs of the feature amount, and creates the decision tree by searching an internal node and the feature amount, in which the evaluation value becomes maximum. Thereby, the decision tree by taking account of the calculation costs as well as the classification accuracy can be created. A data classification unit 5 inputs the data to be classified, performs the branching by calculating the feature amount for each root node and internal node by using the decision tree created by the decision tree learning unit 3 and outputs a category indicated by a reached leaf node as a classification result. Thereby, since the decision tree in which the calculation costs also are taken into account is used, the processing time is shortened. <P>COPYRIGHT: (C)2011,JPO&INPIT</p> |
