Logistic model tree

In computer science, a logistic model tree (LMT) is an algorithm for supervised learning tasks which is combined with linear logistic regression and tree induction. Logistic model trees may consist of any combination of binary decisions based on numeric attributes and logistic regression models at leaf nodes.

The logistic model tree is generated by first constructing a logistic model at the root using the LogitBoost algorithm to iteratively fit simple linear-regression functions. LogitBoost uses simple regression functions as base functions, and performs forward stagewise fitting of additive symmetric logistic-regression models, based on maximum-likelihood calculations (Friedman, et al. 2000) and the C4.5 splitting criterion (Sumner, et al. 2005), thereby selecting variables that best predict the target variable. The LMT model is then extended at child nodes by using LogitBoost. Splitting at a node continues in this fashion until no additional useful split can be found. The LMT tree is then pruned using the CART crossvalidation-based pruning algorithm.

See also

• C4.5 algorithm