Decision trees are well-known and established models for classification and regression. In this paper, we propose multi-path decision tree algorithm (MPDT). Different from traditional decision tree where the path of each record is deterministic and exclusive, a record can trace several paths simultaneously in multi-path decision tree so that it has the effect of ensemble classifiers with only one classifier. Local class information gain is the value of class information (entropy or Gini, etc) given the value of an attribute relative to class information unsupervised. We examine the MPDT on a random selection of 26 benchmark data sets from the UCI repository and compared it with Bagging, AdaBoost and C4.5. The results note that MPDT has better performance.
展开▼
