statistic classification method of data using one-sided purity splitting criteria for classification trees in data mining

摘要

PURPOSE: A method for splitting a pure interest node on classification trees for data mining is provided to offer a new splitting method to form a high interpretability classification tree within a range not damaging accuracy largely. CONSTITUTION: An independent variable and a threshold are selected as a splitting standard. Child nodes are decided according to the splitting standard. It is judged as the splitting process termination if an observation value of the child node is fallen under a constant value. If the observation value of the child node is above the constant value, the splitting process is continued.