Classification Trees With Unbiased Multiway Splits

By Hyunjoong Kim, Wei-Yin Loh. 2001

440 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Information Systems, Plant Science

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

In Plant Science

Citations in the corpus

Unbiased Recursive Partitioning: A Conditional Inference Framework (2006, methodology): “Further methodological developments within this framework include the incorporation of a linear discriminant analysis model within each node of a tree (Kim and Loh 2003) and multiway splits (CRUISE, Kim and Loh 2001).”
Multivariate decision trees using linear discriminants and tabu search (2003, methodology): “Algorithms that apply linear discriminant analysis to the decision tree building process were developed in [17], [26], [31], and [32].”
Model-Based Recursive Partitioning (2008, methodology): “The logistic regression-based MOB trees are compared with logistic model trees (LMT, Landwehr, Hall, and Frank 2005) as well as various tree-based algorithms with constant models in the nodes: QUEST (Loh and Shih 1997), CRUISE (Kim and Loh 2001), the J4.”
Classification and regression trees (2012, background): “Probably the most important flaw concerns a selection bias toward covariates with many possible splits (White and Liu 1994) or missing values (Kim and Loh 2001).”
Classification and Regression via Integer Optimization (2007, methodology): “…(via MATLAB’s® neural network toolbox), SVM (via SVMfu4) (Rifkin 2002), and classification trees with univariate and multivariate partition rules (via CRUISE5—a classification tree software with both univariate and multivariate partitioning (Kim and Loh 2000, Loh and Shih 1997, Kim and Loh 2001)).”
A new variable importance measure for random forests with missing data (2012, methodology): “To overcome this problem, several unbiased tree algorithms have been suggested (see, e.g., White and Liu 1994; Kim and Loh 2001; Dobra and Gehrke 2001; Hothorn et al. 2006; Strobl et al. 2007).”
Classification tree analysis using TARGET (2008, methodology): “For example, we might measure terminal node purity as the number of misclassified observations in a terminal node, and the tree fitness criterion could simply be the sum of the terminal node purities, i.e., the total number of observations misclassified by the tree.”
Recent advances in decision trees: an updated survey (2022, methodology): “Among DTs that employ LDA family methods, this paper highlights FACT (Loh and Vanichsetakul 1988), CRUISE (Kim and Loh 2001), QUEST (Loh and Shih 1997), and GUIDE (Loh 2009).”
Understanding Random Forests: From Theory to Practice (2014, background): “been known for long in decision trees [Kononenko, 1995; Kim and Loh, 2001; Hothorn et al., 2006], we argue that it is not the primary cause for the bias observed here.”
Classification and regression trees (2011, neutral): “10 , 11 , 28 The computational speed of C4.5 is almost always the fastest, whereas RPART can be fast or extremely slow, with the latter occurring when Y takes more than two values and there are unordered variables taking many values (owing to its coding, the RPART software may give incorrect…”