## Global discretization of continuous attributes as preprocessing for machine learning (1996)

### Cached

### Download Links

- [kuscholarworks.ku.edu]
- [kuscholarworks.ku.edu]
- DBLP

### Other Repositories/Bibliography

Venue: | International Journal of Approximate Reasoning |

Citations: | 49 - 3 self |

### BibTeX

@INPROCEEDINGS{Chmielewski96globaldiscretization,

author = {Michal R. Chmielewski and Jerzy W. Grzymala-busse},

title = {Global discretization of continuous attributes as preprocessing for machine learning},

booktitle = {International Journal of Approximate Reasoning},

year = {1996},

pages = {294--301}

}

### Years of Citing Articles

### OpenURL

### Abstract

Abstract. Real-life data usually are presented in databases by real numbers. On the other hand, most inductive learning methods require small number of attribute values. Thus it is necessary to convert input data sets with continuous attributes into input data sets with discrete attributes. Methods of discretization restricted to single continuous attributes will be called local, while methods that simultaneously convert all continuous attributes will be called global. In this paper, a method of transforming any local discretization method into a global one is presented. A global discretization method, based on cluster analysis, is presented and compared experimentally with three known local methods, transformed into global. Experiments include ten-fold cross validation and leaving-one-out methods for ten real-life data sets.

### Citations

5430 |
C4.5 – Programs for Machine Learning
- Quinlan
- 1993
(Show Context)
Citation Context ... "best" breakpoints which together with the domain boundary points induce the desired intervals (Minimal Class Entropy Method) was suggested in [7]. A similar method of discretization is used in C4.5 =-=[13]-=-. The class information entropy of the partition induced by a break-point q is defined as E (A, q; U) = |S1 |U| Ent(S1) + |S2| |U| Ent(S2). 3sThe break point q for which E(A, q; U) is minimal among al... |

4443 |
Classification and Regression Trees
- Breiman, Friedman, et al.
- 1984
(Show Context)
Citation Context ...land. The data set glass, representing glass types, has been created by B. German, Central Research Establishment, Home Office Forensic Science Service, Canada. The data set waveform, as described in =-=[1]-=-, represents three types of waves. The data set image created in 1990 by the Vision Group, University of Massachusetts, represents image features: brickface, sky, foliage, cement, window, path, and 11... |

519 |
Rough sets
- Pawlak
- 1982
(Show Context)
Citation Context ...ent. We need a measure of consistency for inconsistent data sets. Our measure, called a level of consistency, is based on rough set theory, a tool to deal with uncertainty, introduced by Z. Pawlak in =-=[12]-=-. Let U denote the set of all examplessof the data set. Let P denote a nonempty subset of the set of all variables, i.e., attributes and a decision. Obviously, set P defines an equivalence relation ℘ ... |

271 |
Computer Systems that Learn: Classification and Prediction Methods from Statistics
- Weiss, Kulikowski
- 1991
(Show Context)
Citation Context ...riterion of the discretization method is the accuracy rate. A complete discussion on how to evaluate the error rate (and hence the accuracy rate) of a rule set induced from a data set is contained in =-=[14]-=-. We have used the following cross validation guidelines in computing the accuracy rate: • If the number of examples was less than 100, the leaving-one-out method was used to estimate the accuracy rat... |

167 | On changing continuous attributes into ordered discrete attributes - Catlett - 1991 |

165 |
On the handling of continuous-valued attributes in decision tree generation
- Fayyad, Irani
- 1992
(Show Context)
Citation Context ... with class-entropy as a criterion to evaluate a list of "best" breakpoints which together with the domain boundary points induce the desired intervals (Minimal Class Entropy Method) was suggested in =-=[7]-=-. A similar method of discretization is used in C4.5 [13]. The class information entropy of the partition induced by a break-point q is defined as E (A, q; U) = |S1 |U| Ent(S1) + |S2| |U| Ent(S2). 3sT... |

84 |
LERS – a system for learning from examples based on rough sets
- Grzymala-Busse
- 1992
(Show Context)
Citation Context ...amples Module, version 2) to induce rules from the discretized data sets. LEM2 is a typical member of the family of learning algorithms in that it finds a minimal discriminant description, see, e.g., =-=[4, 8]-=-. The most important performance criterion of the discretization method is the accuracy rate. A complete discussion on how to evaluate the error rate (and hence the accuracy rate) of a rule set induce... |

35 |
Synthesizing Statistical Knowledge from Incomplete Mixed-Mode Data
- Wang, Chiu
- 1987
(Show Context)
Citation Context ...ipal disadvantage of this method, however, is the repetition of the learning process until the final performance level is reached. A discretization based on maximal marginal entropy was introduced in =-=[15]-=-. This process involves partitioning the domain of the continuous attribute such that the sample frequency in each interval is approximately equal and is called Equal Frequency per Interval Method. Th... |

33 | Information discovery through hierarchical maximum entropy discretization and synthesis - Chin, Wong, et al. - 1991 |

17 | Statistical Analysis for Decision Making - Hamburg - 1983 |

11 |
Determination of quantization intervals in rule based model for dynamic systems
- Chan, Batur, et al.
- 1991
(Show Context)
Citation Context ... continuous attribute is partitioning its domain into equal width intervals is called Equal Interval Width Method. A method of attribute discretization through adaptive discretization was proposed in =-=[3]-=-. The domain of an attribute is first partitioned into two equal width intervals and a learning system is run to induce rules. Then, the quality of the rules is tested by evaluating rule performance. ... |

6 | Quantization of numerical sensor data for inductive learning - Pao, Bozma - 1986 |

5 |
On the attribute redundancy and the learning programs ID3, PRISM, and LEM2
- Chan, Grzymala-Busse
- 1991
(Show Context)
Citation Context ...amples Module, version 2) to induce rules from the discretized data sets. LEM2 is a typical member of the family of learning algorithms in that it finds a minimal discriminant description, see, e.g., =-=[4, 8]-=-. The most important performance criterion of the discretization method is the accuracy rate. A complete discussion on how to evaluate the error rate (and hence the accuracy rate) of a rule set induce... |

2 | Probabilistic approach to design algorithm generation in the case of continuous condition attributes - Lenarcik, Z - 1992 |