Similarity classifier using similarity measure derived from Yu's norms in classification of medical data sets.

摘要

A new approach using a similarity measure based on Yu's norms is presented for the detection of erythemato-squamous diseases, diabetes, breast cancer, lung cancer and lymphography. The domain contains records of patients with known diagnoses. The results are very promising with all data sets and (in conclusion, can be drawn that) a similarity model derived from Yu's norms could be used for the diagnosis of patients taking into consideration the error rate. A similarity classifier derived from Yu's norms was used to detect the six erythemato-squamous diseases when 34 features defining six disease indications were used as inputs. The results confirmed that the proposed model has potential in detecting the erythemato-squamous diseases. The similarity model derived from Yu's norms achieved an accuracy rate (97.8%) which was higher than that of the stand-alone neural network model or the ANFIS model suggested in Ubeyli and Guler [Automatic detection of erythemato-squamous diseases using adaptive neuro-fuzzyinference systems, Comput. Biol. Med. 35 (2005) 421-433] or the similarity model based on Lukasiewicz similarity [Luukka and Leppalampi, Similarity classifier with generalized mean applied to medical data, Comput. Biol. Med. 36 (2006) 1026-1040]. With PIMA Indian diabetes, the detection model has an error rate of about 24% which is much better than the overall rate of 33% for diabetes. Also, a classifier was applied to the lung cancer data set and the results were to my knowledge better than before. When the lung cancer data were preprocessed with an entropy minimization technique and the classifier with similarity based on Yu's norm was applied, 99.99% accuracy was achieved. The use of this preprocessing method enhanced the results over 30%. In lymphography, entropy minimization also enhanced the results remarkably and 86.2% accuracy was achieved.