Improving the Prediction Accuracy of Liver Disorder Disease with Oversampling

The complexity of liver makes it easily affected by disease of disorder. So diagnosing liver disorder disease is a high interest to data miners, and decision trees have been useful data mining tools to diagnose the disease, but the accuracy of decision trees has been limited due to insufficient data. In order to generate more accurate decision trees for liver disorder disease this paper suggests a method based on over-sampling in minor classes to compensate the insufficiency of data effectively. Experiments were done with two representative algorithms of decision trees, C4.5 and CART, and a data set, ‘BUPA liver disorder’, and showed the validity of the method. Key-Words: biased sampling, liver disorder disease, classification.