## Learning Decision Lists Using Homogeneous Rules (1994)

### Cached

### Download Links

- [ftp.cs.washington.edu]
- [ftp.cs.washington.edu]
- [www.research.ibm.com]
- DBLP

### Other Repositories/Bibliography

Venue: | In AAAI-94 |

Citations: | 39 - 2 self |

### BibTeX

@INPROCEEDINGS{Segal94learningdecision,

author = {Richard Segal and Oren Etzioni},

title = {Learning Decision Lists Using Homogeneous Rules},

booktitle = {In AAAI-94},

year = {1994},

pages = {619--625},

publisher = {AAAI press}

}

### Years of Citing Articles

### OpenURL

### Abstract

A decision list is an ordered list of conjunctive rules (Rivest 1987). Inductive algorithms such as AQ and CN2 learn decision lists incrementally, one rule at a time. Such algorithms face the rule overlap problem --- the classification accuracy of the decision list depends on the overlap between the learned rules. Thus, even though the rules are learned in isolation, they can only be evaluated in concert. Existing algorithms solve this problem by adopting a greedy, iterative structure. Once a rule is learned, the training examples that match the rule are removed from the training set. We propose a novel solution to the problem: composing decision lists from homogeneous rules, rules whose classification accuracy does not change with their position in the decision list. We prove that the problem of finding a maximally accurate decision list can be reduced to the problem of finding maximally accurate homogeneous rules. We report on the performance of our algorithm on data sets from the ...

### Citations

3480 | Induction of decision trees
- Quinlan
- 1986
(Show Context)
Citation Context ... 2 and on all the data sets from the MONK's competition (Thrun et al. 1991). The results are shown in Table 3. For comparison, the results for the IND (Buntine and Caruana 1991) implementation of C4 (=-=Quinlan 1986-=-) are also included. The results for the UCI data sets are averaged over 10 iterations. Each iteration randomly splits the available data into 70% for training and 30% for testing. The MONK's problems... |

210 |
Scientific explanation and the causal structure of the world
- Salmon
- 1984
(Show Context)
Citation Context ...is probabilistic. Our main contribution is a solution to the overlap problem that is both theoretically justified and practical. We borrow the notion of homogeneity from the philosophical literature (=-=Salmon 1984-=-) to solve the overlap problem in learning decision lists. Informally, a homogeneous rule is one whose accuracy does not change with its position in the decision list. Formally, let E denote the unive... |

203 |
Boolean feature discovery in empirical learning
- Pagallo, Haussler
- 1986
(Show Context)
Citation Context ...s algorithm is impractical because the number of decision lists is doublyexponential in the number of attributes. Many existing algorithms (e.g., (Michalski 1969; Clark and Niblett 1989; Rivest 1987; =-=Pagallo and Haussler 1990)) learn d-=-ecision lists incrementally by searching the space of conjunctive rules for "good" rules and then combining the rules to form a decision list. Such algorithms face the problem of rule overla... |

177 | The MONK's Problems - A Performance Comparison of Dierent Learning Algorithms - Thrun, Bala, et al. - 1991 |

78 |
Rule induction using information theory
- Smyth, Goodman
- 1990
(Show Context)
Citation Context ...earch on a complex real-world domain. Brute's running time on the two data sets was less than 3 CPU minutes on a SPARC-10 workstation. 3 nSeveral other systems have used depth-bounded search. ITRULE (=-=Smyth and Goodman 1991-=-), like 3 We previously reported Brute's running time as 33 CPU minutes. We have since added additional pruning axioms that significantly improve Brute's efficiency. Brute, uses depth-bounded search t... |

69 |
On the quasi-minimal solution of the general covering problem
- Michalski
(Show Context)
Citation Context ... of decision lists and output the best one found. This algorithm is impractical because the number of decision lists is doublyexponential in the number of attributes. Many existing algorithms (e.g., (=-=Michalski 1969; Clark an-=-d Niblett 1989; Rivest 1987; Pagallo and Haussler 1990)) learn decision lists incrementally by searching the space of conjunctive rules for "good" rules and then combining the rules to form ... |

65 | Efficiently inducing determinations: A complete and systematic search algorithm that uses optimal pruning - Schlimmer - 1993 |

58 | Exploring the decision forest: An empirical investigation of Occamâ€™s razor in decision tree induction - Murphy, Pazzani - 1994 |

40 | Representation design and brute-force induction in a boeing manufacturing domain
- Riddle, Segal, et al.
- 1994
(Show Context)
Citation Context ... when a pure depth-bounded search to the desired depth is too costly. The basic ideas behind BruteDL apply equally well to heuristic search. Related Work BruteDL builds on our previous work on Brute (=-=Riddle et al. 1994-=-). Brute uses a depth-bounded search of the space of conjunctive rules to find accurate predictive rules. We tested Brute on two data sets from a Boeing manufacturing domain. The first data set has 1,... |

30 | Introduction to ind and recursive partitioning
- Buntine
- 1991
(Show Context)
Citation Context ... data sets from the UCI repository (Murphy 1994) 2 and on all the data sets from the MONK's competition (Thrun et al. 1991). The results are shown in Table 3. For comparison, the results for the IND (=-=Buntine and Caruana 1991-=-) implementation of C4 (Quinlan 1986) are also included. The results for the UCI data sets are averaged over 10 iterations. Each iteration randomly splits the available data into 70% for training and ... |

20 | Systematic search for categorical attribute-value data-driven machine learning
- Webb
- 1993
(Show Context)
Citation Context ...hms (e.g., (Michalski 1969; Clark and Niblett 1989)). These algorithms share Rivest's iterative structure but use a beam search to find the best rule according to a scoring function. The OPUS system (=-=Webb 1993-=-) extends CN2 to use depth-bounded search but retains the same iterative structure. As a result, poor rule choices at the beginning of the list can significantly reduce the accuracy of the decision li... |

7 |
Learning decision trees
- Rivest
- 1987
(Show Context)
Citation Context ... of Computer Science and Engineering University of Washington Seattle, WA 98195 fsegal, etzionig@cs.washington.edu Appears in AAAI-94 Abstract A decision list is an ordered list of conjunctive rules (=-=Rivest 1987-=-). Inductive algorithms such as AQ and CN2 learn decision lists incrementally, one rule at a time. Such algorithms face the rule overlap problems--- the classification accuracy of the decision list de... |

2 | Constructing decision trees in noisy domains - Irvine - 1994 |