P. Clark and R. Boswell, "Rule induction with cn2: Some recents improvements," in Machine Learning: Proc. 5th Eur. Conf., 1991, pp. 151--163.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....specify a different minimum item support for each item. Different rules may need to satisfy different minimum supports, depending on what items are in the rules. Besides the measures we have mentioned, some other measures for a rule include gain [FMM 96] entropy gain [MFM 98, Mor98] and laplace [CB91, Web95]. 2.9. Derive High Confidence Rules In some situations, users are interested in finding rules with high confidence but not necessarily high support. The algorithms we discussed before, such as Apriori, cannot handle this problem because they are support based algorithms. There are some studies ....

P. Clark and P. Boswell, "Rule Induction with CN2: Some Recent Improvements," Proceedings of the European Machine Learning Conference, Porto, Portugal, March 1991, pp. 151-163.

....by consistency gain. Indeed, in the examples shown below, the two methods provide identical rule rankings. If there are more than two decision classes, the entropy terms are summed. Nonetheless, the above comments regarding no consideration of rule completeness remain true. A later version of CN2 [4] offered a new rule quality formula based on a Laplace error estimate. This formula is closely tied to a rule s consistency level, while completeness still plays a minimal role. IREP s formula for rule evaluation [8] is simply: p N n) P N) 11) RIPPER, as was mentioned in Section 2, uses ....

Clark, P. and Boswell, R., "Rule Induction with CN2: Some Recent Improvements," in Kodratoff, Y. (ed.), Proceedings of the Fifth European Working Session on Learning (EWSL-91), Berlin: Springer-Verlag, pp. 151-163, 1991.

....are used to generate classification rules from class labeled examples that are described by a set of numerical (e.g. 1,2,4) nominal (e.g. black, white) or continuous attributes. Some of the inductive ML algorithms, like AQ algorithm [20] 15] CLIP algorithms [5] 6] or CN2 algorithm [8] [9], can handle only numerical or nominal data, while some others can handle continuous attributes but still perform better with discretevalued attributes [1] 17] This drawback can be overcome by using a discretization algorithm as a front end for a machine learning algorithm. Discretization ....

P. Clark and R. Boswell, "Rule Induction with CN2: Some Recent Improvements", Lecture Notes in Artificial Intelligence, Proceedings of the European Working Session on Learning, SpringerVerlag, 1991.

.... dataset, the calculations for each attribute are also used in the dataset description (the lowest part of Table 1) 3 Experiments Three different propositional classification algorithms were used in the experiments: tree learning algorithm C4.5 [8] rule learning algorithm CN2 with m estimate [4, 5] and k nearest neighbour (k NN) algorithm [10] These algorithms were used both for base level and meta level learning. For base level learning, they were applied to twenty datasets from the UCI Repository of Machine Learning Databases and Domain Theories [7] For meta level learning, the three ....

Clark, P. and Boswell, R. (1991) Rule induction with CN2: Some recent improvements. In Proceedings of the Fifth European Working Session on Learning, pages 151--163. Springer.

....MDTs and MLC4.5 are described in Section 3. The performance of MDTs is evaluated on a collection of twenty one data sets. MDTs are used to combine classi ers generated by ve base level learning algorithms: two treelearning algorithms C4.5 [17] and LTree [11] the rule learning algorithm CN2 [7], the k nearest neighbor (k NN) algorithm [20] and a modi cation of the naive Bayes algorithm [12] In the experiments, we compare the performance of stacking with MDTs to the performance of stacking with ODTs. We also compare MDTs with two voting schemes and two other stacking approaches. ....

....as the number of (internal and leaf) nodes, as a measure of simplicity: the smaller the tree, the simpler it is. 4. 2 Base Level Algorithms Five learning algorithms have been used in the base level experiments: two tree learning algorithms C4.5 [17] and LTree [11] the rule learning algorithm CN2 [7], the k nearest neighbor (k NN) algorithm [20] and a modi cation of the naive Bayes algorithm [12] All algorithms have been used with their default parameters settings. The output of each base level classi er for each example in the test set consist of at least two components: the predicted ....

Clark, P. and Boswell, R. (1991) Rule induction with CN2: Some recent improvements. In Proceedings of the Fifth European Working Session on Learning: 151-163. Springer-Verlag.

....trees. The description of the method is given in Section 2. The performance of the proposed method is evaluated on a collection of twenty one data sets. We combine models generated by ve learning algorithms: two tree learning algorithms C4.5 [12] and LTree [8] the rule learning algorithm CN2 [4], the k nearest neighbor (k NN) algorithm [14] and a modi cation of the naive Bayes algorithm [9] In the experiments, we compare the performance of stacking with MDTs to the performance of stacking with ODTs. We also compare MDTs with two voting schemes. Section 3 reports on the experimental ....

....the UCI Repository of Machine Learning Databases and Domain Theories [11] These data sets have been widely used in other comparative studies. Five learning algorithms were used in the base level experiments: two treelearning algorithms C4.5 [12] and LTree [8] the rule learning algorithm CN2 [4], the k nearest neighbor (k NN) algorithm [14] and a modi cation of the naive Bayes algorithm [9] All algorithms were used with their default settings. We used ve di erent algorithms for combining classi ers. Two of them are voting schemes, and three are based on stacking: P VOTE is a simple ....

Clark, P. and Boswell, R. (1991) Rule induction with CN2: Some recent improvements. In Proceedings of the Fifth European Working Session on Learning: 151-163. Springer-Werlag.

....addressing the above kinds of problems. The mathematical foundation of LAD is in discrete mathematics, with a special emphasis on the theory of Boolean functions. Patterns are the key building blocks in LAD [21, 13] as well as in many other rule induction algorithms (such as C4.5rules [23] CN2 [10, 9], AQ17 HCI [26] RISE [14] RIPPER [11] and SLIPPER [12] Since a typical data set has an exceedingly large number of patterns, all these algorithms are limited to the consideration of small subsets of patterns. In most algorithms, the choice of such a subset of patterns is not explicitly ....

P. Clark and R. Boswell. "Rule induction with CN2: Some recent improvements", in Machine Learning -- Proceedings of the Fifth European Conference (EWSL-91) (ed. Y Kodrato#), 151--163. Berlin: Springer-Verlag, 1991. http://www.cs.utexas.edu/users/pclark/papers/newcn.ps

....7 2.1 Database Names Database names are sometimes a source of confusion in case studies. For example, when a case study such as [121] refers to the soybean database, which of the four soybean databases located in ftp: ics.uci.edu pub machine learning databases soybean is meant Or when [37] refers to the thyroid database, which of the eleven thyroid disease databases in ftp: ics.uci.edu pub machine learning databases soybean thyroid disease is meant These two case studies provide just two examples of imprecise or incomplete specification of data a cursory search of the machine ....

....88, 105] to be especially desirable since rules are easily understood by humans. The CN2 induction algorithm incorporates ideas from Quinlan s ID3 decision tree algorithm and from Michalski s AQ. CN2 generates ordered (ocn2) or unordered (ucn2) lists of classification rules and is described in [35, 36, 37]. The CN2 software used in this study was obtained from David Aha [23] Software for CN2 is also available from the Turing Institute [24] The c45 rule algorithm uses the decision tree constructed by c45 tree to build production rules. Quinlan describes the method used to transform a decision tree ....

[Article contains additional citation context not shown here]

Peter Clark and Robin Boswell (1991). Rule Induction with CN2: Some Recent Improvements. Machine Learning--EWSL-91: European Working Session on Learning in Lecture Notes in Artificial Intelligence Vol. 482, Springer-Verlag. 151-163.

....MDTs and MLC4.5 are described in Section 3. The performance of MDTs is evaluated on a collection of twenty one data sets. MDTs are used to combine classi ers generated by ve base level learning algorithms: two treelearning algorithms C4.5 [17] and LTree [10] the rule learning algorithm CN2 [6], the k nearest neighbor (k NN) algorithm [20] and a modi cation of the naive Bayes algorithm [11] In the experiments, we compare the performance of stacking with MDTs to the performance of stacking with ODTs. We also compare MDTs with two voting schemes. Finally, we compare MDTs to boosting and ....

.... 0.16 wine 7.83 1.63 6.95 1.51 2.81 0.74 2.36 0.34 1.46 0.30 Average 14.67 0.89 14.78 0.77 14.07 0.77 13.89 0.89 14.30 0. 39 13 Five learning algorithms were used in the base level experiments: two tree learning algorithms C4.5 [17] and LTree [10] the rule learning algorithm CN2 [6], the k nearest neighbor (k NN) algorithm [20] and a modi cation of the naive Bayes algorithm [11] All algorithms were used with their default settings. The output of each base level classi er for each example in the test set consist of at least two components: the predicted class and the class ....

Clark, P. and Boswell, R. (1991) Rule induction with CN2: Some recent improvements. In Proceedings of the Fifth European Working Session on Learning: 151-163. Springer-Verlag.

....the leaves of the tree. Relational learning systems applied include ICL (De Raedt and Van Laer 1995) which induces classification rules, SRT (Kramer 1996) and TILDE (Blockeel and De Raedt 1998) The latter are capable of inducing both classification and regression trees. ICL is an upgrade of CN2 (Clark and Boswell 1991) to first order logic, TILDE is an upgrade of C4.5, and SRT is an upgrade of CART (Breiman et al. 1984) TILDE cannot construct linear models in the leaves of its trees; SRT can. Finally, FFOIL (Quinlan 1996) was also applied to the classification version of the problem. It used a representation ....

Clark, P. and Boswell, R. 1991. Rule induction with CN2: some recent improvements. In Y. Kodratoff, ed., Proceedings of the Fifth European Working Session on Learning, Lecture Notes in Artificial Intelligence, vol. 482, pp. 151-163. Springer.

....is called the base classifier. The most popular class binarization rule is the unordered or one against all class binarization, where one takes each class in turn and learns binary concepts that discriminate this class from all other classes. It has been independently proposed for rule learning (Clark Boswell, 1991), neural networks (Anand et al. 1995) and support vector machines (Cortes Vapnik, 1995) Definition 2.2 (unordered class binarization) The unordered class binarization transforms a class problem into 2 class problems. These are constructed by using the examples of class as the positive ....

....binarization) The unordered class binarization transforms a class problem into 2 class problems. These are constructed by using the examples of class as the positive examples and the examples of classes , as the negative examples. The name unordered originates from Clark and Boswell (1991), who proposed this approach as an alternative to the decision list learning approach that was originally used in CN2 (Clark Niblett, 1989) In other fields, the strategy has different names, but as our main concern is rule learning, we stick with the terminology used there. The rule learning ....

Clark, P., & Boswell, R. (1991). Rule induction with CN2: Some recent improvements. Proceedings of the 5th European Working Session on Learning (EWSL-91) (pp. 151--163). Porto, Portugal: Springer-Verlag.

....algorithms [106, 132] aimed at the automatic extraction of rules or decision trees from data. Early machine learning systems, dealing with real world data which may be erroneous (noisy) and incomplete, include CART [18] Quinlan s extensions to ID3 [133] ASSISTANT [17, 24] AQ [109] and CN2 [29, 28]. The C4.5 system [135] is an e#cient and probably the most popular machine learning system of the nineties. inference engine knowledge base protocols, guidelines, etc. data mining temporal abstraction visualization user interface . ....

....domain of early diagnosis of rheumatic diseases. 30 If then rule induction was studied previously by Michalski [108] and implemented in a series of AQ algorithms, e.g. the AQ15 system which was also applied for the analysis of medical data [109] Here we describe the rule induction system CN2 [29, 28] which is among the best known of if then rule learners, capable also of handling imperfect noisy data. Like the AQ algorithms, CN2 also uses the covering approach to construct a set of rules for each possible class c i in turn: when rules for class c i are being constructed, examples of this ....

Clark, P., Boswell, R., "Rule induction with CN2: Some recent improvements." In: Proc. Fifth European Working Session on Learning, Springer, 1991, pp. 151--163.

....will be described in a forthcoming book [ Michie et al. 1994 ] 2 Previous Comparison Studies A number of criticisms can be leveled at previous comparative studies; most concern algorithms and datasets. Many comparisons have been intra subject comparisons, e.g. within symbolic learning [ Clark and Boswell, 1991, Sammut, 1988, Quinlan et al. 1986, Aha, 1992 ] within statistics [ Cherkaoui and Cleroux, 1991, Remme et al. 1980 ] and within neural networks [ Huang et al. 1991, Fahlman, 1991, Xu et al. 1991, Ersoy and Hong, 1991 ] There have been fewer, 3 but still many, inter subject studies ....

....and rather good fit . 3 Algorithms and Datasets The algorithms that we have chosen are representative of the established state of the art in the respective fields. They can be divided into four categories: ffl Symbolic classifiers, including C4.5 [ Quinlan, 1987a ] NewID, AC 2 , CN2 [ Clark and Boswell, 1991 ] ITrule [ Goodman and Smyth, 1989 ] CART [ Breiman et al. 1984 ] and INDCART; ffl Statistical classifiers, including k nearest neighbor, naive Bayes, linear discriminant (Discrim) quadratic discriminant (Quadra) and logistic regression (LogReg) Hand, 1981 ] ffl Modern ....

P. Clark and R. Boswell. 1991 Rule induction with cn2: some recent improvements. EWSL '91, pp 151--163, Porto, Portugal. Berlin: Springer-Verlag.

....class note that with zero examples the probability of each class is 1 C . This may or may not be desirable for a speci c problem; however, practitioners have found the Laplace correction worthwhile. To our knowledge, the Laplace correction was introduced in machine learning by Niblett (1987) Clark and Boswell (1991) incorporated it into the CN2 rule learner, and its use is now widespread. For decision tree learning the Laplace correction has been used by certain researchers and practitioners (Pazzani et al. 1994; Bradford, Kunz, Kohavi, Brunk, Brodley, 1998; Provost et al. 1998; Bauer Kohavi, 1999; ....

Clark, P., & Boswell, R. (1991). Rule induction with CN2: Some recent improvements. In Proceedings of the Sixth European Working Session on Learning, pp. 151-163 Porto, Portugal. Springer.

No context found.

P. Clark and R. Boswell, "Rule induction with cn2: Some recents improvements," in Machine Learning: Proc. 5th Eur. Conf., 1991, pp. 151--163.

No context found.

Clark, P., & Boswell, R. (1991). Rule induction with CN2: Some recent improvements. Proceedings of the 5th European Working Session on Learning (EWSL-91) (pp. 151-- 163). Porto, Portugal: Springer-Verlag.

No context found.

Clark, P. & Boswell, R. (1991). Rule induction with CN2: Some recent improvements. In Proc. Fifth European Working Session on Learning (pp. 151--163). Springer.

No context found.

CLARK,P.AND BOSWELL, R. 1991. Rule induction with CN2: Some recent improvements. In Proc. Fifth European Working Session on Learning. Springer, Berlin, 151--163.

No context found.

P. Clark, R. Boswell, "Rule Induction with CN2 : Some Recent Improvements", Proceedings of the 6th European Working Sessions on Learning, pp 151-163.

No context found.

P. Clark and R. Boswell, "Rule induction with cn2: Some recents improvements," in Machine Learning: Proc. 5th Eur. Conf., 1991, pp. 151--163.

No context found.

CLARK, P. -- BOSWELL, R. 1991. Rule induction with CN2: Some recent improvements. Proceeding Fifth European Working Session on Learning, pages 151 -- 163, Springer, Berlin.

No context found.

CLARK, P. -- BOSWELL, R. 1991. Rule induction with CN2: Some recent improvements. Proceeding Fifth European Working Session on Learning, pages 151 -- 163, Springer, Berlin.

No context found.

P. Clark and R. Boswell, "Rule Induction with CN2: Some Recent Improvements," Proc. European Working Session on Learning, 1991.

No context found.

P. Clark and R. Boswell, "Rule induction with cn2: Some recents improvements," in Machine Learning: Proc. 5th Eur. Conf., 1991, pp. 151--163.

No context found.

Clark, P., & Boswell, R. (1991). Rule induction with CN2: Some recent improvements. Proceedings of the 5th European Working Session on Learning (EWSL-91) (pp. 151--

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