Generating decision trees with discriminants and employing the same in data classification

摘要

At least a portion of a decision tree structure is generated from one or more multidimensional data objects by representing data associated with one or more of the data objects as a node, determining a condition for dividing the data at the node into at least two subsequent nodes based on a discriminant measure which maximises the separation between classes associated with the data, and dividing the data according to the condition. The multidimensional objects may be data records including feature variables and class variables and the method comprises splitting a decision tree, recursively, such that the greatest amount of separation among the class values of the data is achieved. The discriminant measure is preferably determined in accordance with Fisher's discriminant technique and the data is divided at a split plane determined to be perpendicular to a direction determined according to said technique and where an entropy measure is substantially optimised, as determined in accordance with a gini index.