J. R. Quinlan, "Generating production rules from decision trees," in Proceedings of the 10th International Joint Conference on Artificial Intelligence, pp. 304--307, 1987.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....which covers all the cases not covered by other rules. These steps do require further assessment of the quality of the generated rules and thus need further scanning of the data. As the pruning step may need to be repeated several times the time spent on this data scanning may be substantial. See [63] for implementation of the rule induction from decision trees. The basic algorithm for the construction of the decision tree consists of repeated assessment of the purity of the nodes in each partition and decisions on which partition to split into subpartitions and how, see Algorithm 6. ....

J.R. Quinlan. Generating production rules from decision trees. In Proc. of IJCAI--87, pages 304--307, 1987.

....[12, 9] also known as the replication problem. The fragmentation problem has been attacked in different ways: by constructing compound features at every tree node [12, 18] by reducing the number of possible partitions [5, 16] and by using alternative concept representations, e.g. sets of rules [15], decision graphs [6, 11] SE Trees [22] decision lists [12, 21] Nonetheless, no clear solution has emerged. Appeared in the 9th European Conference on Machine Learning (1997) Lecture Notes in Artificial Intelligence, Vol. XXX, Springer Verlag, Heidelberg, pp 312 326. Our analysis of the ....

....of all partial hypotheses (step 1) 5 Experiments 5.1 The Learning Systems Used for Testing We use C4.5trees [16] to represent a decision tree inducer where each splitting function tests on a single feature. The importance of global data analysis is underlined in modified versions of C4.5rules [15, 16], as explained in Sect. 5.3. For a decision tree inducer with multiple feature tests, we developed a new version of the LFC system [18] the new version is called DALI [26] Dynamic Adaptive Lookahead Induction) In both DALI and LFC , a splitting function is defined as the conjunction of several ....

Quinlan, J. R.: Generating Production Rules from Decision Trees. In Proceedings of the 10th International Joint Conference on Artificial Intelligence, (1987) 304--307

....technique to reduce the number of pairs of records which needs to be considered, 6 3. automatic induction of a decision tree model, pruning to avoid overfitting and transformation of the induced model to production rules for increasing the performance of the similarity detection process [Qui87], and 4. uncertainty reasoning (based on certainty factors and fuzzy set theory) for masking the errors in the predictive behavior of the model related to both record sampling and the specific values selected by the inductive algorithm for splitting up the domain of each one of the features. 3 ....

....converted into the consequent ( then part ) of the rule. As a side effect, we try to increase the accuracy of the induced model and, at the same time, decrease its complexity and size for efficiency reasons. For further improvement of the quality of the rule set, we adopt the method proposed in [Qui87] where a 2 Theta 2 contingency table is built for each of the rules after removing one condition from its antecedent. Confidence or certainty factors offer a simple tool for representing uncertainty. In expert systems, they have been used to denote confidence that stated facts and rules do indeed ....

J. R. Quinlan. Generating Production Rules From Decision Trees. In Proc. 10 Tenth International Joint Conference on Artificial Intelligence, pages 304--307, 1987.

....cannot be resolved in a straightforward way when rule sets are merged. Each rule that is created will have associated with it a measure of its goodness which is based on its accuracy and the number and type of examples it covers. We are using a normalized version of Quinlan s certainty factor [Quinlan, 1987, Provost and Hennessy, 1996] to determine the accuracy of a rule R over an example set E as: acc(R, E) TP 0.5) TP pFP) 1) where TP is the number of true positives examples covered by R when applied to E, FP is the number of false positives caused by R when applied to E, and p is the ....

J.R. Quinlan. Generating production rules from decision trees. In Proceedings of IJCAI87, pages 304-307, 1987.

....implementing them listed below. 3.1 Decision Trees Decision trees are perhaps the most widely studied inductive learning models in the machine learning community. The literature abounds with papers proposing new models or variations of existing models and case studies using decision trees ([14, 21, 22, 25, 30, 34, 22 40, 43, 49, 50, 51, 53, 89, 93, 98, 99, 100, 101, 102, 104, 105, 106, 107, 109, 110, 111, 112, 113, 114, 118, 120, 123, 126, 129, 130, 131, 133, 134, 136]) For this case study, we use decision tree software from Quinlan and Buntine. Quinlan introduces decision trees and illustrates the use of his C4.5 software for decision trees (c4.5tree) and production rules derived therefrom (c4.5rule) in [105] Several decision tree algorithms (cart, id3, c4, ....

....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 into rules in [100, 105]. The c45 rule software is available with Quinlan s book [105] 3.6 Neural Networks The perceptron is the oldest of all inductive learners having first been described by Rosenblatt [116, 117] and Hebb [60] in the 60s and in 1949 respectively. Minsky and Papert discuss the perceptron and its ....

J.R. Quinlan (1987). Generating Production Rules from Decision Trees. Proceedings of the 10th International Joint Conference on Artificial Intelligence. Morgan Kaufmann Publishers, San Mateo, CA. 304-307.

....or pruning the complex tree. The process of growing is usually governed by some heuristic functions such as binary split which recursively partition the data. CART uses the Gini index to measure the impurity at a node [45] and then chooses the split to maximize the reduction in impurity; ID3 [112] and C4.5 [118] use a measure of information gain; and the CHAID decision tree algorithm [74] uses a splitting criterion based on chi square test statistic [pp. 36 37, 144] On the pruning process, there are two approaches depending on whether or not a testing data set is used to estimate the ....

....pruning strategy. In general, the learning algorithms start with an initial rule set and iteratively improve the rule set using heuristic techniques [30] Another common approach is to firstly generate a tree, then convert the tree into an equivalent rule set and finally prune unnecessary rules [112]. Strength and Limitation 30 Systems that learn sets of rules have a number of desirable properties. Rule sets are relatively easy for people to understand [20] they are the most comprehensible classification models. Also rule learning systems usually outperform decision tree learners [103, ....

[Article contains additional citation context not shown here]

Quinlan J.R., "Generating Production Rules From Decision Trees", Int Joint Conf. on Artificial Intelligence, pp.304-307, 1987.

....data mining tools, Addriaans Zantinge (1996) have been appeared in literature, able to discover effectively patterns and regularities contained in data records of large databases. Furthermore, most of these tools and methodologies are capable of presenting the results in an intelligible form [Quinlan (1987)] while they can use them also to make predictions about key properties [Fayyad et al. 1996) In fact, the majority of such systems consists a mature version of machine learning algorithms and tools, widely appeared in literature during the 80 s [Michalski et al. 1983) Mitchell (1997) ....

QUINLAN, J. R. (1987). "Generating Production Rules from Decision Trees". In: IJCAI Proceedings, Morgan Kaufmann, Milan, 304-307.

....rule is in general more easily interpretable by the user than a long rule. The rule pruning procedure is performed for each ant as soon as the ant completes the construction of its rule. The search strategy of our rule pruning procedure is very similar to the rule pruningproceduresuggestedbyQuinlan (1987), although the rule quality criterion used in the two procedures are very different from each other. The basic idea is to iteratively remove one term at a time from the rule while this process improves the quality of the rule. A more detailed description is as follows. In the first iteration one ....

Quinlan, J.R. (1987). Generating production rules from decision trees. Proc. 1987 Int. Joint Conf. on Artif. Intel.(IJCAI), 304-307.

.... developed in applied statistics in the 1960 s (e.g. MS63] but did not receive much attention within statistics until the publication of the pioneering work on CART [BFOS94] Quinlan independently popularized the use of trees in machine learning with his ID3 and C4.5 family of algorithms [Qui87, Qui93]. Both the CART and ID3 C4.5 approaches share many common ideas, resulting in quite similar tree learning algorithms. The statistical work on trees typically emphasizes parameter estimation and tree selection aspects of the problem, while more recent work on trees in data mining has emphasized ....

Quinlan, J. R. (1987) Generating production rules from decision trees, Proceedings of the Tenth International Joint Conference on Artificial Intelligence, San Mateo, CA: Morgan Kaufmann, 304--307. 26

....are due to A. Srinivasan and J. Catlett for contributing their databases. We thank D. Aha for comments on a draft of this paper, and the reviewers for many valuable suggestions to improve it. 35 A Description for Algorithms A. 1 Symbolic algorithms C4.5, NewID and AC 2 C4.5 [ Quinlan, 1987a, Quinlan, 1987b ] NewID, AC 2 are descendant of ID3 [ Quinlan, 1986 ] They select an attribute for splitting a dataset into subsets according to the result of a test conducted on the attribute. This set is used to form the root of the tree and the chosen attribute divides it into subsets. Each subset is ....

J.R. Quinlan. 1987. Generating production rules from decision trees. International Joint Conference on Artificial Intelligence, pp 304--307, Milan.

....a series of if then rules and tests each of them against a set of data. In practice, RL is often used to find interesting individual rules. However, the set of rules learned by RL forms a disjunctive class description, which can be optimized with standard techniques as described, for example, by Quinlan (1987). RL performs a straightforward, general to specific search of the space of rules defined by conjunctions of attribute value pairs (features) The goal of RL s search is to find rules that satisfy user defined criteria. In particular, in the experiments below RL searches for rules that satisfy ....

Quinlan, J. (1987). Generating production rules from decision trees. Proceedings of the Tenth International Joint Conference on Artificial Intelligence (pp. 304--307). San Mateo, CA: Morgan Kaufmann.

....algorithm can take advantage of inter association rule constraints to nd association rules eciently. 2. BACKGROUND Early approaches to identifying interesting rules from data were dominated by attempts to form small sets of rules for accurate classi cation of further previously unsighted data [9, 7, 13]. For the most part, borrowing from an elegant characterization of mining optimized rules by Bayardo and Agrawal [3] this activity can be characterized as follows: A training set is a nite set of records where each record is an element to which we apply Boolean predicates called conditions. ....

J. R. Quinlan. Generating production rules from decision trees. In IJCAI 87: Proceedings of the Tenth International Joint Conference on Articial Intelligence, pages 304-307, Los Altos, 1987. Morgan Kaufmann.

....be learned by a learning system. For example, a decision tree learning algorithm searches only the space of finite trees that carry out axis orthogonal splits. Decision trees [Hunt, Marin, and Stone, 1966; Quinlan, 1983; Breiman et al. 1984; Quinlan, 1993a] and production rules [Michalski, 1978; Quinlan, 1987a] are two commonly used theory description languages in supervised learning. Other theory description languages include neural networks [Rumelhart et al. 1986; Hinton, 1989] linear discriminant functions [James, 1985] and instance based methods [Aha, 1990; Aha, Kibler, and Albert, 1991] An ....

....is constructed from a path of a tree, it is natural to expect that the new attribute is good at discriminating examples of the class labeled by the leaf of the path from examples of other classes. However, some conditions in a path may be irrelevant to the class labeled by the leaf of the path [Quinlan, 1987a] Consequently, new attributes created directly from paths by using the fixed path based strategy may contain some irrelevant conditions. On the other hand, a system called C4.5rules [Quinlan, 1987a; 1993a] transforms decision trees into production rules. In the process, irrelevant or unimportant ....

[Article contains additional citation context not shown here]

J.R. Quinlan, Generating production rules from decision trees. Proceedings of the Tenth International Joint Conference on Artificial Intelligence, San Mateo, CA: Morgan Kaufmann, 304-307.

....rule where the intersection of all the internal nodes form the body of the rule and the class in the leaf is the head. Association Rules can be discovered as paths on Decision Trees (DTs) that represent the data. Indeed, association rules have been constructed with the help of decision trees (cf. [25]) Decision trees have been treated extensively in the machine learning community (cf. 24, 26, 25] as a mechanism for representing decisions about a specified goal, based on the values of attributes of the data records. The most popular algorithm for DT construction is perhaps Quinlan s ID3 ....

....in the leaf is the head. Association Rules can be discovered as paths on Decision Trees (DTs) that represent the data. Indeed, association rules have been constructed with the help of decision trees (cf. 25] Decision trees have been treated extensively in the machine learning community (cf. [24, 26, 25]) as a mechanism for representing decisions about a specified goal, based on the values of attributes of the data records. The most popular algorithm for DT construction is perhaps Quinlan s ID3 [24] and its improvements, as grouped in the set of programs C4.5 [26] The more recent DT construction ....

J. R. Quinlan. Generating production rules from decision trees. In IJCAI-87, pages 304--7, San Mateo, CA, 1987.

....like the set of rules generated by a rule induction classifier: the tests that appear in the way from the root of a decision tree to a leaf, can be translated to a rule s IF part, the predicted class of the leaf also in the THEN part appears. Some approaches in this way can be found in Quinlan [35]. Some problems that may be overcome by the rule induction paradigm are: generation of simple rules when noise is present to avoid the overfitting and efficient rule generation when using large databases. Some works in this paradigm are Clark and Nibblet [6] Michalski et al. 28] and Niblet et ....

J.R. Quinlan (1987): "Generating Production Rules from Decision Trees", J. McDermott editor, IJCAI-87, 304-307. San Francisco, CA. Morgan Kaufmann.

....Several lines of investigation of decision trees over finite information systems are known. These are test theory [5, 14, 16, 17, 31, 37, 38] theory of information systems and rough set theory [23, 24, 30] theory of questionnaires [25, 26] theory of decision tables [9] machine learning [27, 28, 29], search theory [1, 35] The terminology and methods of test theory, the groundwork for which was laid by [5, 37] will be used in the present paper for study of decision tree complexity as well as methods of rough set theory created in [23, 24, 30] 160 M.Moshkov Unimprovable Upper Bounds on ....

J.R. Quinlan, Generating production rules from decision trees, Proc. of the 10th Int. Joint Conf. on AI, pp. 304-307 (1987).

....count of the number of instances satisfying its argument. Since we are often interested in nding small rules, it is wise to use a statistical correction to this frequencybased estimate to reduce the number of spurious rules. Programs that mine conjunctive rules often use either Yates correction (Quinlan, 1987) or the Laplace correction (Segal Etzioni, 1994) to obtain a better estimate of con dence. A common way to use con dence as an interestingness criterion is to specify a con dence threshold below which rules are not interesting. Support, also called coverage, is an estimate of p(A 1 : A ....

.... eciency To demonstrate the e ect of breadth rst marker propagation, we replaced matching with bfmp in the rule space search algorithm (bfmp RL) As above, we use a w beam search and a rule complexity limit to restrict the search space of rules, use con dence with the correction described by (Quinlan, 1987) to evaluate rule interestingness. The algorithm accepts a rule if its con dence is above a user de ned threshold. 27 To test our rst analytical result that, even without hierarchical background knowledge, breadth rst marker propagation is more ecient than conventional matching as the number ....

Quinlan, J. R. (1987). Generating production rules from decision trees. In Proceedings of the Tenth International Joint Conference on Articial Intelligence, pp. 304-307. Morgan Kaufmann.

....trees. This is a well known technique in machine learning field of AI, widely used in general purpose systems for classification tasks. Such trees will be inferred using information extracted from a POS annotated corpus and they can be directly interpreted in terms of classification rules ( [Qui87] ) or constraints. In order to test the appropriateness of decision trees to the concerning problem we have restricted, by now, the focus of our study to the learning of part of speech tagging rules, which are the most important part at the base of constraint grammars (constraint rules related to ....

Quinlan, J.R. Generating Production Rules from Decision Trees. In Proceedings of the Tenth International Joint Conference on Artificial Intelligence, pp. 304-307. Milan, Italy: Morgan Kaufmann, 1987.

....dimensions, even with high performance workstations, timing considerations make it necessary to emphasize only the most promising directions. Five classification methods were tried: Linear Discriminant [8] k Nearest Neighbor [4] Neural Network [9] Tree Classification [2] and Rule Induction [3,10,11,12,15]. These methods were applied to the smaller N1 population. All error rates were measured by test cases obtained by randomly holding out 1=3of the sample cases. No method achieved an error rate better than 38 . The following are the results for the k Error Rate 1 .52 5 .49 11 .48 25 .48 Table ....

J.R. Quinlan. Generating Production Rules From Decision Trees. In Proceedings of the Tenth IJCAI, pages 304--307, 1987.

....high dimensions, even with high performance workstations, timing considerations make it necessary to emphasize only the most promising directions. Five classification methods were tried: Linear Discriminant[7] k Nearest Neighbor [4] Neural Network [8] Tree Classification [2] and Rule Induction [3, 9, 10, 11, 15]. These methods were applied to the smaller N1 population. All error rates were measured by test cases obtained by randomly holding out 1=3 of the sample cases. No method achieved an error rate better than 38 . The following are the results for the (reduced size) N1 population. k Nearest ....

J.R. Quinlan. Generating Production Rules From Decision Trees. In Proceedings of the Tenth IJCAI, pages 304--307, 1987.

....and by discarding redundant clauses. A literal is irrelevant if, after it has been eliminated, the clause does not cover any new negative examples. A clause is redundant if it is covered by some more general clause. Postprocessing is especially e#ective when transforming decision trees into rules #Quinlan 1987#. In step showresult the obtained DHDBRules are stored as a ConceptDescription in a #le. 4 Learning nonrecursive de#nitions of relations: A comparison with FOIL This section discusses the performance of LINUS on four learning tasks taken from machine learning literature, whichwere used by ....

Quinlan, J.R. #1987# Generating production rules from decision trees. Proc. Int. Joint Conference on Arti#cial Intelligence, IJCAI 87, Milano, Italy.

No context found.

J. R. Quinlan, "Generating production rules from decision trees," in Proceedings of the 10th International Joint Conference on Artificial Intelligence, pp. 304--307, 1987.

No context found.

J.R. Quinlan. Generating production rules from decision trees. In Proc. of IJCAI--87, pages 304--307, 1987.

No context found.

J. R. Quinlan. Generating production rules from decision trees. In Proceedings of the Tenth International Joint conference on Artificial intelligence, pages 304--307, Milan, Italy, Aug 1987.

No context found.

Quinlan, J.R. (1987). Generating Production Rules from Decision Trees. Proceedings of the 4th International Machine Learning Workshop (pp. 31-37), San Mateo, CA, Morgan Kaufmann

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