PROBLEM TO BE SOLVED: To generate a contracted feature which is more compact and more accurate than a feature used for general supervised learning.SOLUTION: In an original feature importance calculation unit 141, by using a base model 22 constructed through learning from correct answer data by a base model construction unit 12 and unanalyzed data 24, importance indicating the influence of each original feature extracted from the unanalyzed data on a maximum likelihood output of the based model 22 with respect to input of the unanalyzed data 24 is calculated for each of plural original features. In an original feature selection unit 142, original features having importance of 0 are excluded, and the remaining original features are selected. In an original feature fusion unit 143, from a set of the selected original features, original features having the same importance are combined as one contracted feature to generate a contracted feature set. In an original feature importance addition unit 144, a feature related to the importance of an original feature is generated and added to the contracted feature set.
展开▼