Visual Analytics - Decision Trees (1)

The construction of decision trees is a supervised machine learning technique, i.e., it is based on carefully annotated ground truth data. In this lecture, we discuss how decision trees are constructed, why they are useful to „explain“ data and how to interactively explore decision trees. We just give an overview of algorithms. We focus more on the parameters used as input and on the properties of the results. Decision tree construction algorithms have stopping criteria and are compared with respect to goodness criteria. Accuracy of the classification provided by a decision tree and the complexity of the tree are essential goodness criteria. Lower complexity is preferred to avoid overfitting to the training data. Large decision trees are also undesirable because high acquisition costs may be caused if too many attributes are needed. Examples from medicine and finance data are discussed. Interaction is essential; experts may integrate their knowledge to prune, refine, or modify decision trees. In this first lecture, we restrict to basic visualization techniques, such as node-link diagrams, treemaps, and icicle plots. Chapters: 00:00 - Outline and Introduction 21:52 - Decision Trees for Classification 38:25 - Goodness Measures 54:35 - Reducing DT Complexity 57:30 - Visualization