R. Kerber. Chimerge: Discretization of numeric attributes. In Proceedings of AAAI-92, 10th Conference of the American Association for Arti cial Intelligence, pages 123-128, San Jose, US, 1998.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....the two tiles that are active for a given point. point in that n dimensional space would fall into just one bin. Identifying a small set of landmark numerical values known as feature discretization to help classification is also a well studied question within machine learning [Cat91, Ker92, FI93, DKS95, KS96, FW99] Feature discretization centers on the issue of finding good split points in a decision tree. As such, the main idea is to find the fewest and the most indicative values that can help make a good prediction. Learning methods such as C4.5 would normally have to consider a ....

R. Kerber. Chi-Merge: Discretization of numeric attributes. In Proceedings of the 10th National Conference on Artificial Intelligence, pages 123-- 128, Menlo Park, CA, 1992. AAAI Press/MIT Press.

....highest probability given an instance. It is plausible that it is less important to form intervals dominated by a single class for naive Bayes classifiers than for decision trees or decision rules. Thus discretization methods that pursue pure intervals (containing instances with the same class) [1, 5, 10, 11, 14, 15, 19, 29] might not suit naiveBayes classifiers. Besides, naive Bayes classifiers deem attributes conditionally independent of each other and do not use attribute combinations as predictors. There is no need to calculate the joint probabilities of multiple attribute values. Thus discretization methods that ....

....X i , suppose there are n training instances for which the value of X i is known, the minimum and maximum value are v min and v max respectively, each discretization method first sorts the values into ascending order. The methods then di#er as follows. 4. 1 Equal Width Discretization (EWD) EWD [5, 9, 19] divides the number line between v min and v max into k intervals of equal width. Thus the intervals have width w = v max v min ) k and the cut points are at v min w, v min 2w, v min (k 1)w. k is a user predefined parameter and is set as 10 in our experiments. 4.2 Equal ....

[Article contains additional citation context not shown here]

Kerber, R. Chimerge: Discretization for numeric attributes. In National Conference on Artificial Intelligence (1992), AAAI Press, pp. 123--128.

....for each interval The usual method chooses the mean of the values that fall on this interval. Due to the influence of outliers some authors prefer to use the median. 2.2 Other systems We now briefly describe some of the existing discretization systems. Unsupervised methods The ChiMerge system [9] begins by placing each observed real value into its own interval and proceeds by using the test to determine when adjacent intervals should be merged. The StatDisc [10] method also uses statistical tests as a means of determining discretization intervals. This is also a bottom up method that ....

Kerber R. Chimerge: discretization of numeric attributes. In Proceedings of the 10 National Conference on Artificial Intelligence. MIT Press, 1992.

....the expected value of that cell. In this case, the expected value is ,where is the total frequency in row, is the total frequency in column , and is the total of all cells in the table. During tree construction, attribute partitions are selected using an approach suggested by Kass [5] and Kerber [6]. For each attribute, a contingency table is constructed with a row for each class value and a column for eachof attribute values every possible value for discrete attributes or every unique interval for discretized continuous attributes. Then, the pair of columns in the table with the least ....

Randy Kerber. Chimerge: Discretization of numeric attributes. In . MIT Press, 1992.

....computed as w krj = 1ifw kj (# r 1 ,# r ] 0 otherwise (8) among all the possible replacements, the one chosen maximizes e#ectiveness. Di#erent techniques for discretizing a continuous attribute have been proposed in the field of machine learning. These include 1R [5] ChiMerge and Chi2 [6], plus several entropy based algorithms [1, 2, 7, 10 12] In this paper, we adopt avery simple discretization algorithm based on (multiclass) information gain, which is defined in terms of the well known (multiclass) entropy measure [9] for reasons discussed later in this section, we take to ....

R. Kerber. Chimerge: Discretization of numeric attributes. In Proceedings of AAAI-92, 10th Conference of the American Association for Artificial Intelligence, pages 123--128, San Jose, US, 1998.

....computed as w krj = 1ifw kj (# r 1 ,# r ] 0 otherwise (8) among all the possible replacements, the one chosen maximizes e#ectiveness. Di#erent techniques for discretizing a continuous attribute have been proposed in the field of machine learning. These include 1R [5] ChiMerge and Chi2 [6], plus several entropy based algorithms [1, 2, 7, 10 12] In this paper, we adopt avery simple discretization algorithm based on (multiclass) information gain, which is defined in terms of the well known (multiclass) entropy measure [9] for reasons discussed later in this section, we take to ....

R. Kerber. Chimerge: Discretization of numeric attributes. In Proceedings of AAAI-92, 10th Conference of the American Association for Artificial Intelligence, pages 123--128, San Jose, US, 1998.

....clustering effect, that is, the number of discretized clusters are the least. The discretized continuous attribute will be regarded as a new category attribute, and used together with previous category attributes for rule extraction. The discretization is done by a modified version of ChiMerge [13]. The instances are sorted according to the value of the attribute being discretized. Initial clusters are formed through regarding each unique value as a cluster. The Z 2 value of adjacent clusters is computed and the pairs of adjacent clusters with the lowest Z 2 value are merged. In Kerber s ....

....are sorted according to the value of the attribute being discretized. Initial clusters are formed through regarding each unique value as a cluster. The Z 2 value of adjacent clusters is computed and the pairs of adjacent clusters with the lowest Z 2 value are merged. In Kerber s original algorithm [13], merging continues until all pairs of clusters that have Z 2 values exceeding a user defined parameter Z2 threshold. Instead of setting a fixed threshold value as a stopping condition for merging, we continue the merging as long as there is no instances that belong to different classes are ....

Kerber R. Chi-Merge: Discretization of Numeric Attributes. In: Proceedings of the 9th National Conference on Artificial Intelligence, Menlo Park, CA: AAAI Press, 1992, 123-128.

....discretized clusters is the least one that is enough for extracting valid rules. The discretized continuous attribute is regarded as a new category attribute, and used together with existing category attributes for rule extraction. Here the discretization is done by a modified version of ChiMerge [34]. The examples are sorted according to the value of the attribute to be discretized. Initial clusters are formed through regarding each unique value as a cluster. The Z 2 value of adjacent clusters is computed and the pairs of adjacent clusters with the lowest Z 2 value are merged. In Kerber s ....

....are sorted according to the value of the attribute to be discretized. Initial clusters are formed through regarding each unique value as a cluster. The Z 2 value of adjacent clusters is computed and the pairs of adjacent clusters with the lowest Z 2 value are merged. In Kerber s original algorithm [34], merging continues until all pairs of clusters that have Z 2 values exceeding a user defined parameter Z2 threshold. Instead of setting a fixed threshold value as a stopping condition for merging, we continue the merging as long as there are no examples that belong to different classes are ....

R. Kerber, Chi-Merge: Discretization of Numeric Attributes. Proc. of the 9th National Conference on Artificial Intelligence, San Jose, CA, 123-128, 1992.

....that the proposed algorithm always generates rules from categorical attributes. Only when new rules can not be generated, the algorithm resorts to the continuous attribute whose number of discretized intervals are the least. The discretization is performed by a modified version of ChiMerge [15]. The instances are sorted according to the values of the attribute being discretized. Initial intervals are formed by making each unique value as an interval. The value of the adjacent intervals is computed and the pairs of adjacent intervals with the lowest value are merged. In ....

....sorted according to the values of the attribute being discretized. Initial intervals are formed by making each unique value as an interval. The value of the adjacent intervals is computed and the pairs of adjacent intervals with the lowest value are merged. In Kerber s original algorithm [15], merging continues until the values of all pairs of intervals exceed a user defined parameter, i.e. threshold. Instead of setting a fixed threshold as the stopping criterion for merging, the proposed algorithm continues the merging as long as there are instances that belong to ....

R. Kerber, "Chi-Merge: discretization of numeric attributes", In Proc. AAAI-92, San Jose, CA, pp.123-128, 1992. TABLE II COMPARISON OF THE CONCISENESS OF THE GENERATED RULES Number of rules Average antecedents Maximum antecedents Minimum antecedents

....attribute will be regarded as Z. H. Zhou et al. Extracting Symbolic Rules from Trained Neural Network Ensembles 5 a new categorical attribute and used along with previous categorical attributes for successive rule extraction. The discretization is performed by a modified version of ChiMerge [21]. The instances are sorted according to the value of the attribute being discretized. Initial clusters are formed by making each unique value as a cluster. The # 2 value of adjacent clusters is computed and the pairs of adjacent clusters with the lowest # 2 value are merged. In Kerber s ....

....are sorted according to the value of the attribute being discretized. Initial clusters are formed by making each unique value as a cluster. The # 2 value of adjacent clusters is computed and the pairs of adjacent clusters with the lowest # 2 value are merged. In Kerber s original algorithm [21], merging continues until the # 2 values of all pairs of clusters exceed a user defined parameter # 2 threshold. Instead of setting a fixed threshold as the stopping criterion for merging, REFNE continues the merging as long as there is no instances that belong to di#erent classes assigned ....

R. Kerber, Chi-Merge: Discretization of numeric attributes, in: Proceedings of the 9th National Conference on Artificial Intelligence, San Jose, CA, 1992, pp.123-128.

...., are not i discrete. However, all machine learning models in this study require that the y be discrete. i Therefore the output classes of the 6 databases (auto, house, import, lrs, mache, and machp) 28 in this study with continuous output attributes are discretized using an equal width method [32, 69, 128]. The discretized output attributes are used for both the machine learning and connectionist learning models even though the connectionist models could have used continuous values for output classes. This was done to eliminate any performance differences which may have resulted from using discrete ....

....learning models are unable to handle continuous input attributes. In this study, CN2 is unable, in its current incarnation, to handle continuous attributes. Thus, before presentation to CN2, all continuous valued input attributes in all database are discretized using the Kerber s Chi Merge method [69]. Note that even though continuous valued input attributes are discretized before presentation to the CN2 software, the comments made above regarding the composition of the training and testing sets and presentation order still apply, that is, the composition and order of the training and testing ....

Randy Kerber (1992). ChiMerge: Discretization of Numeric Attributes. AAAI-92 Proceedings of the 10th National Conference on Artificial Intelligence. AAAI Press/MIT Press, Cambridge, Massachusetts. 123-127.

....stage requires an instance to be able to follow more than one branch of a node ending up, maybe, in more than one leaf. Classification is then straightforward by averaging the instance classes available at all the leaves reached by an instance. An algorithm that uses the 2 metric, ChiMerge [11], has been used to discretize continuous attributes. ChiMerge employs a 2 related threshold to find the best possible points to split a continuum of values. The value for 2 threshold is determined by selecting a desired significance level and then using a table to obtain the ....

Kerber, R., Chimerge: Discretization of numeric attributes. In proceedings of the 10 th National Conference on Artificial Intelligence, San Jose, CA. MIT Press. 123-128, 1992.

....by a set of numerical, nominal, or continuous attributes. Many existing inductive ML algorithms are designed expressly for handling numerical or nominal data, while some algorithms perform better with discrete valued attributes despite the fact that they can also handle continuous attributes [1] [9]. This drawback can be overcome by using a discretization algorithm as a front end for the learning algorithm. Discretization is a process of transforming a continuous attribute values into a finite number of intervals and associating with each interval a discrete, numerical value. The usual ....

....interdependence. The representative algorithms are: maximum entropy [16] Patterson Niblett algorithm [11] which is built in as a front end into a decision trees algorithm [13] and other information gain or entropy based algorithms [7] 18] statisticsbased algorithms like ChiMerge [9] or Chi2 [10] class attribute interdependency algorithms like CADD algorithm [2] and clustering based algorithms like K means discretization algorithm [15] Discretization should significantly reduce the number of possible values of the continuous attribute since large number of possible ....

Kerber R.: ChiMerge: Discretization of Numeric Attributes, Proc. AAAI-91, 9th International Conference on Artificial Intelligence, pp.123-128, 1992

....D. Bay able) and do not consider interactions with other features. For example, Fayyad and Irani (1993) recursively split an attribute to minimize the class entropy. They use a minimum description length criterion to determine when to stop. Other algorithms in this category include: ChiMerge (Kerber, 1992), Chi2 (Liu Setiono, 1995) error based discretization (Kohavi Sahami, 1996) and many others. Dougherty, Kohavi and Sahami (1995) and Zighed et. al (1999) provide good overviews of many of the classical discretization algorithms. Elomaa and Rousu (1999) examined methods for finding an optimal ....

....to speedup methods such as windowing, sampling, and limiting the depth of the search (Provost Kolluri, 1999) This will speed up MVD accordingly. 4.2. Relation to Other Discretization Approaches Our bottom up merging process is similar to other discretization algorithms such as ChiMerge (Kerber, 1992) and Chi2 (Liu Setiono, 1995) They divide the data into intervals and then merge them on the basis of a chi square test checking for independence of interval membership and class. Our work differs in our merging criteria as we require that the two intervals have substantially different ....

Kerber R (1992) Chimerge: Discretization of numeric attributes. Proceedings of the Tenth National Conference on Artificial Intelligence, pp. 123--128.

No context found.

R. Kerber. Chimerge: Discretization of numeric attributes. In Proceedings of AAAI-92, 10th Conference of the American Association for Arti cial Intelligence, pages 123-128, San Jose, US, 1998.

No context found.

Kerber, R.: Chimerge: Discretization of numeric attributes. In: Proc. of AAAI-92, San Jose, CA (1992) 123--128

No context found.

R. Kerber. Chimerge: Discretization of numeric attributes. In Proceedings of the Tenth National Conference on Artificial Intelligence, pages 123--128. AAAI Press/MIT Press, 1992.

No context found.

R. Kerber, Chi-Merge: Discretization of numeric attributes, in: Proceedings of the 9th National Conference on Artificial Intelligence, San Jose, CA, 1992, pp.123-128.

No context found.

R. Kerber, "ChiMerge: Discretization of Numeric Attributes," Proc. Ninth Int'l Conf. Artificial Intelligence (AAAI-91), pp. 123-128, 1992.

No context found.

R. Kerber, ChiMerge: discretization of numeric attributes, in: Proceedings of Ninth International Conference on Artificial Intelligence (AAAI-91), 1992, pp. 123--128.

No context found.

Kerber R. (1992). Chimerge: Discretization of numeric attributes. In Proceedings of the Tenth Nat. Conference on Artificial Intelligence, 123-128. MIT Press.

No context found.

R. Kerber, Chi-Merge: Discretization of numeric attributes, in: Proceedings of the 9th National Conference on Artificial Intelligence, San Jose, CA, 1992, pp.123-128.

No context found.

Kerber, R. Chimerge: Discretization for numeric attributes. In National Conference on Artificial Intelligence (1992), AAAI Press, pp. 123--128.

No context found.

R. Kerber, ChiMerge: discretization of numeric attributes, Proceedings of the 9th International Conference on Artificial Intelligence (AAAI-92), The AAAI Press, MIT, 1992, pp. 123 -- 128.

No context found.

R. Kerber, "Chimerge: Discretization of Numeric Attributes," AAAI-92, Proc. Ninth National Conf. on Artificial Intelligence, AAAI Press/The MIT Press, 1992, pp. 123-- 128. 20

First 50 documents  Next 50

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC