Optimization based decision trees for multi-modal problems

Multi-modal problems are amongst the most difficult-to-handle classification problems, especially for traditional statistical techniques. Multi-modal problems arise when each class region can occupy disjoint areas in feature space. Backpropagation neural networks and decision tree classifiers (DTCs) can typically handle multi-modal problems. We introduce a decision tree based on clustering and linear programming and compare its performance to CART on a number of data sets from the literature, including several sets that exhibit clear multi-modal structure