Better decision tree induction for limited data sets of liver disease

Abstract

Decision trees can be very useful data mining tools for human experts to diagnose the disease, because the knowledge structure is represented in tree shape. But we may not get satisfactory decision tree, if we do not have enough number of consistent instances in the data sets. Recently two kinds of relatively small data sets of liver disorder from America and India are available, so in order to generate more accurate and useful decision trees for the disease this paper suggests appropriate sampling for the data instances that are in the class of higher error rate. Experiments with the two public domain data sets and a representative decision tree algorithm, C4.5, shows very successful results. © 2012 Springer-Verlag.