Generating Production Rules from Decision Trees

By J. Ross Quinlan. 1987

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

Explainable Artificial Intelligence (XAI): Concepts, Taxonomies, Opportunities and Challenges toward Responsible AI (2019, positive): “Rule based learners are clearly transparent models that have been often used to explain complex models by generating rules that explain their predictions [126,127,190,191] .”
A Survey of Methods for Explaining Black Box Models (2018, background): “presents a possible outcome [79]. The paths from the root to the leaves represent the classication rules. Indeed, a decision tree can be linearized into a set of decision rules with the if-then form [77,78,26]: if condition 1 ^condition 2 ^condition 3 then outcome: Here, the outcome corresponds to t”
A Genetic Algorithm for Interpretable Model Extraction from Decision Tree Ensembles (2017, methodology): “A more straight-forward technique is proposed by Quinlan [5] which translates the decision trees in production rules that are much easier to simplify than the trees themselves.”
A SAT-Based Approach to Learn Explainable Decision Sets (2018, background): “As a result, most algorithms for learning rule lists or sets are heuristic [7, 8, 33, 34], being in general efficient to run, but providing essentially no guarantees in terms of the quality of the computed rules.”
Foundations of Rule Learning (2012, background): “Quinlan (1993) has noted that even pruned decision trees may be too cumbersome, complex, and inscrutable to provide insight into the domain at hand and has consequently devised procedures for simplifying decision trees into pruned production rule sets (Quinlan, 1987a, 1993).”
Generating Accurate Rule Sets Without Global Optimization (1998, methodology): “RIPPER implements this strategy using reduced errorpruning (Quinlan, 1987b), which sets some trainingdata aside to determine when to drop the tail of arule, and incorporates a heuristic based on the mini-mum description length principle as stopping criterion.”
Trading Accuracy for Simplicity in Decision Trees (1994, unknown): “These systems, such as ID3 (Quinlan, 1979, 1983), C4 (Quinlan, 1987), CART (Breiman, et al.”
Machine Learning, Neural and Statistical Classification (2009, methodology): “A compromise is to use decision tree induction to build an initial tree and then derive rules from the tree thus transforming an efficient but opaque representation into a transparent one (Quinlan, 1987b).”
Efficient Filtering on Hidden Document Streams (2014, background): “" Many other types of classifiers like decision trees (Quinlan 1987) and linear classifiers (Towell and Shavlik 1993; Fung, Sandilya, and Rao 2005) can be easily converted to sets of rules; hence we believe that our modeling is quite general.”
Binary Rule Generation via Hamming Clustering (2002, methodology): “5 [19], [20] and the algorithm RAMP described in [28], [49], which is based on a similar approach.”
Unifying Instance-Based and Rule-Based Induction (1996, background): “…in the literature can be seen as empirical multi-strategy learners, but combining di(cid:11)erent paradigms from RISE's: decision trees and rules (Quinlan, 1987), decision trees and perceptrons (Utgo(cid:11), 1989b), rules and Bayesian classi(cid:12)cation (Smyth, Goodman& Higgins, 1990),…”
Instance-based learning algorithms (2004, background): “We suspect this is the same requirement C4 (Quinlan, 1987), PLS (Rendell, 1988), and back-propagation (Rumelhart et al., 1987) need to ensure that target concepts are learnable.”
Data Mining Techniques (2001, background): “In this case the Laplace estimator c(A ⇒ B) ≈ NA∪B + 1 NA + k , where 1/k is an a priori guess of the confidence, may improve accuracy (Segal and Etioni 1994); an alternative is the Yates correction (Quinlan 1987) for an application to rule induction.”
Learning decision lists (2004, methodology): “In Valiant's original paper (1984), the probability distribution pn was constrained so that only positive examples were shown to the learning algorithm.”
Separate-and-Conquer Rule Learning (1999, background): “Quinlan (1987a) sorts the rules by their advantage, i.e., by the number of examples that would be erroneously classified after deleting the rule.”