A further comparison of splitting rules for decision-tree induction

By Wray L. Buntine, Tim Niblett. 2004

143 citations recorded by the source corpus.

One approach to learning classification rules from examples is to build decision trees. A review and comparison paper by Mingers (Mingers, 1989) looked at the first stage of tree building, which uses a “splitting rule” to grow trees with a greedy recursive partitioning algorithm. That paper considered a number of different measures and experimentally examined their behavior on four domains. The main conclusion was that a random splitting rule does not significantly decrease classificational accuracy. This note suggests an alternative experimental method and presents additional results on further domains. Our results indicate that random splitting leads to increased error. These results are at variance with those presented by Mingers.

Subfields: Artificial Intelligence, Information Systems

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Information Systems

Citations in the corpus

Conservative Q-Improvement: Reinforcement Learning for an Interpretable Decision-Tree Policy (2019, background): “The tree, once constructed, also has useful information about which states are treated identically and which features yield the highest information gain [Buntine and Niblett, 1992; Leroux et al. , 2018].”
Extremely randomized trees (2006, background): “…B-28, B-4000 Belgium e-mail: P.Geurts@ulg.ac.be D. Ernst e-mail: Dernst@ulg.ac.be L. Wehenkel e-mail: L.Wehenkel@ulg.ac.be Springer shown in a more carefully designed experiment that they were actually significantly less accurate than normal ones on many datasets (Buntine and Niblett, 1992).”
An introduction to decision tree modeling (2004, background): “Some additional resources are available concerning the selection of decision tree partitions[8,9].”
Metric Regression Forests for Correspondence Estimation (2015, methodology): “…these suc- 100 cesses, the somewhat arbitrary choice to bootstrap using 101 a PARTS objective, clashes with the experience of several 102 authors Buntine and Niblett (1992), Liu and White (1994), 103 Nowozin (2012) who show that the objective function has 104 a substantial influence on the…”
Multivariate Decision Trees (1995, background): “Breiman, et al. (1984), Quinlan (1986a), Mingers (1989a), Safavian and Landgrebe (1991), Buntine and Niblett (1992) and Fayyad and Irani (1992b) discuss and compare diierent partition merit criteria.”
Subtree Replacement in Decision Tree Simplification (2012, methodology): “, pruning vs grafting), we adopt the standard approach of [7], which resorts to a two-tailed paired t-test of the differences of error rates over the 50 executions at some significance level.”
Separate-and-Conquer Rule Learning (1999, methodology): “In general the search heuristics used in separate-and-conquer rule learning are similar to the heuristics used in other inductive learning algorithms like those discussed in (Mingers 1989; Buntine and Niblett 1992).”
Top-down induction of decision trees classifiers - a survey (2005, background): “Comparative studies of the splitting criteria described above, and others, have been conducted by several researchers during the last thirty years, such as [24], [25], [5], [26], [17], [27], [28], [71], [73].”
Very Simple Classification Rules Perform Well on Most Commonly Used Datasets (1993, methodology): “In addition to these fourteen datasets, the study includes a 2-class version of GL (G2), and, following (Buntine & Niblett, 1992), a version of VOi nw hich the "best" attribute has been deleted (V1).”
Extracting decision trees from trained neural networks (2002, methodology): “The ones applicable to general Neural Networks were computationally expemive, Craven addressed the problem by extracting MorN Decision Trees from trained Neural Network.”