Active Feature Selection Based on a Very Limited Number of Entities

摘要

In data analysis, the necessary data are not always prepared in a database in advance. If the precision of extracted classification knowledge is not sufficient, gathering additional data is sometimes necessary. Practically, if some critical attributes for the classification are missing from the database, it is very important to identify such missing attributes effectively in order to improve the precision. In this paper, we propose a new method to identify the attributes that will improve the precision of Support Vector Classifiers (SVC) based solely on values of candidate attributes of a very limited number of entities. In experiments, we show the incremental addition of attributes by the proposed method effectively improves the precision of SVC using only a very small number of entities.