[ Expanding the Scope of Concept Learning Using Metafeatures

Mohammed Waleed Kadouswaleed@cse.unsw.edu.au School of Computer Science and Engineering, University of New South Wales, Sydney 2052 Australia ]

Abstract:

We present a general automated preprocessing technique called metafeatures. Using metafeatures, the scope of traditional propositional attribute-value learning is expanded to domains that do not normally fit in the propositional model. These are domains that contain instances that have some kind of recurring substructure, such as strokes in handwriting recognition, or local maxima in time series data. Metafeatures are applied to three domains: sign language recognition, ECG classification and Chinese handwriting recognition. Using metafeatures we are able to generate classifiers that are both comprehensible and accurate, producing results that are comparable to hand-crafted feature extraction and in one case comparable to human experts.