FUSINTER: A Method for Discretization of Continuous Attributes

Abstract

In induction graphs methods such as C4.51 or SIPINA2, taking continuous attributes into account needs particular discretization procedures. In this paper, we propose on the one hand, an axiomatic leading to a set of criteria which can be used for continuous attributes discretization, and on the other hand, a method of discretization called FUSINTER. The results obtained by FUSINTER are compared to those obtained by techniques developed by Fayyad and Irani3 and Kerber4 and they have proved better for the majority of the examples studied.

MODL: A Bayes optimal discretization method for continuous attributes

“…-MODL -MDLPC (Fayyad & Irani, 1992) -BalancedGain (Kononenko, Bratko, & Roskar, 1984) -Fusinter (Zighed et al, 1998) -Khiops (Boullé, 2003(Boullé, , 2004…”

“…For the Fusinter criterion, we use the regularization parameters recommended in Zighed et al (1998). The Khiops probability parameter is set to 0.95.…”

“…The chi-square-based criteria (Kass, 1980;Bertier & Bouroche, 1981;Kerber, 1991) focus on the statistical point of view whereas the entropy-based criteria (Catlett, 1991;Quinlan, 1993) focus on the information theoretical point of view. Other criteria such as Gini (Breiman et al, 1984) or Fusinter criterion (Zighed, Rabaseda, & Rakotomalala, 1998) try to find a trade off between information and statistical properties. The Minimum Description Length (MDL) criterion (Fayyad & Irani, 1992) is an original approach that attempts to minimize the total quantity of information both contained in the model and in the exceptions to the model.…”

Research on Water Supply Reservoir Operating Rules Extraction Based on Artificial immune Recognition System

“…Information entropy as the measures of the random signals can be applied to extract the information of operating decision-making hidden in the annual runoff as the hydrologic random variable. In this paper the supervised discrete method, FUSINTER proposed by Zighed (1998) [21], is adopted by AIRS to establish new model, information-entropy based AIRS (IEAIRS), to divide the 45 annual runoff into six intervals with the relation to the AHCs (illustrated in table 5), and the intervals by the FUSINTER are compared with those by the frequencies of the annual runoff (seeing the Fig. 1).…”

Khiops: A Statistical Discretization Method of Continuous Attributes

“…CHAID (Kass, 1980) relies on a discretization method close to ChiMerge (Kerber, 1991). SIPINA takes advantage of the Fusinter criterion (Zighed, 1998) based on measures of uncertainty that are sensitive to sample size. While most discretization methods are employed as a preprocessing step to an induction algorithm, there are still other approaches, similar to the wrapper approach involved in the feature selection field.…”