SOM Classification Method based on Transduction Scheme

摘要

Transductive confidence machines (TCMs) when used in classification problems can provide us with reliability for every classification. Many machine learning algorithms, such as KNN algorithm, etc., have been incorporated with TCM, while there's no SOM classification method based on TCM. Considering properties of SOM map unit, this paper first designs a novel nonconformity measurement and TCM-SOM classification method; and then its classification accuracy that is much more better than that of SOM and is close or even higher than that of TCM-KNN is also proved by UCI machine learning datasets.