In Defense of C4.5: Notes in Learning One-Level Decision Trees

By Tapio Elomaa. 1994

How researchers see this paper

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

Citations in the corpus

Interpretable Machine Learning: Fundamental Principles and 10 Grand Challenges (2021, background): “Elomaa (1994) and Freitas (2014) make the point that “Humans by nature are mentally opposed to too simplistic representations of complex relations.””
Comprehensibility of Classification Trees–Survey Design (2019, background): “Humans by nature are mentally opposed to too simplistic representations of complex relations [7], therefore it is no surprise that empirical studies have shown comprehensibility to be negatively correlated with the complexity (size) of a classifier in at least some cases [1].”
Occam's Two Razors: The Sharp and the Blunt (1998, background): “A similar measure for decision trees would always achieve the Bayes rate (lowest error possible). most, these experiments suggest that the advantage of going to more complex models is small; they do not imply that simpler models are better (Elomaa 1994).”
Comprehensible classification models: a position paper (2014, background): “Second, users are unlikely to accept models that are very simple when the underlying concept being learned by the algorithm is considered complex [20].”
Human Evaluation of Models Built for Interpretability (2019, background): “Allahyari and Lavesson (2011); Elomaa (2017) also ﬁnd that larger models can sometimes be more interpretable.”
Interpretability of machine learning models and representations: an introduction (2016, background): “According to him, following the medical example of [18], experts can be opposed to over-simplistic models.”
Theory and Applications of Agnostic PAC-Learning with Small Decision Trees (1995, background): “[ Elo94 ]. Unfortunately very little information of this type is available at this point, since it is usually too time consuming to compute min for interesting classes of simple hypotheses and common datasets (for an exception see [WGT90] for the case of production rules of length 3 on the dataset )”