Detecting Similarity of Transferring Datasets Based on Features of Classification Rules

In order to transfer mined knowledge for various datasets obtained from transferring situations, it is important to detect not only availability of transferring the knowledge but also detecting their limitations of the transfer. Although most of methods to detect the limitations use performance indices of sets of classifiers such as accuracies of classifier sets, those of each classifier are also useful. Data characterizing techniques have been developed to control learning algorithm selection by using statistical measurements of a dataset. Expanding this framework, we consider a method to reuse objective rule evaluation indices of classification rules such as support, precision, and recall, to measure similarity of different datasets. In this paper, we present a method to characterize given datasets based on objective rule evaluation indices and classification learning algorithms. The experimental results show the method can detect similarity of datasets even if the datasets have totally different attribute sets. This indicates that the limitations of transferring both of classifiers and learning algorithms can be detected as the similarity among datasets by using a learning algorithm.