A Soft Decision Tree
Hung Son Nguyen · 2002
Searching for binary partition of attribute domains is an important task in Data Mining, particularly in decision tree methods. The most important advantage of decision tree methods are based on compactness and clearness of presented knowledge and high accuracy of classification. In case of large data tables, the existing decision tree induction methods often show to be inefficient in both computation and description aspects. The disadvantage of standard decision tree methods is also their instability, i.e., small deviation of data perhaps cause a total change of decision tree. We present the novel “soft discretization” methods using “soft cuts” instead of traditional “crisp” (or sharp) cuts. This new concept allows to generate more compact and stable decision trees with high classification accuracy. We also present an efficient method for soft cut generation from large data bases.