In-depth feature selection for continuous data classification with granule mining

This study proposes theories and algorithms to select features for in-depth groups of data identified as granules, using granule mining techniques to improve the classification performance of continuous data. Global, class-specific, and granule-specific feature selections are performed to observe the significance of granule-specific feature selection when the datasets contain objects that cannot be correctly defined by the given features. Since the granules can be treated as the subclasses of a dataset, the outcomes of this study can be used for selecting features and mining in-depth knowledge in data analysis applications where the data does not contain the subclass information.