Quinlan, J. R. (1987). Decision trees as probabilistic classifiers. Proceedings of the Fourth International Workshop on Machine Learning (pp. 31-37). Irvine, CA: Morgan Kaufmann.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....after this point. Other algorithms are robust up to 80 noise level. 5.2.3.3 Experiments with Increasing Ratio of Missing Values Most of the real world data sets contain missing attribute values. In the literature, some methods are proposed to handle instances containing missing feature values [26, 52, 53, 54, 55]. These methods can be summarized as: 0 20 40 60 80 Missing Value Ratio ( 65.0 67.0 69.0 71.0 73.0 75.0 77.0 79.0 81.0 83.0 85.0 87.0 89.0 91.0 93.0 95.0 FI1 FI2 FI3 FI4 CFP NBC k NNFP k NN Figure 5.3. Accuracy results of the FIL, CFP, NBC, k NN and k NNFP algorithms on ....

J.R. Quinlan, Decision Trees as Probabilistic Classifiers, In Proceedings of Fourth International Workshop on Machine Learning, pp: 31-37, June 1987.

....noise. 5.2.3.3 Experiments with Increasing Level of Missing Values Most of the real world datasets contain missing (unknown) feature values and the percentage of missing values are shown in Table A.1. In order to cope with instances that contain missing values, several methods have been proposed [30, 55, 56, 57, 58]. These methods can be summarized as: ffl Ignoring instances which have unknown feature values. ffl Assuming an additional special value for unknown attribute values. ffl Using probability theory by utilizing information provided by context. ffl Generating additional instances for all possible ....

J.R. Quinlan, Decision Trees as Probabilistic Classifiers, In Proceedings of Fourth International Workshop on Machine Learning, 31--37, June 1987.

....created regions. Figure 2.8 shows an example of the types of planes MSM T might construct. The resulting planes can then be used in the manner of a decision tree to classify new points. MSM T has been shown to learn concepts as well or better than more traditional learning methods such as C4.5 [71, 72] and CART [14] It also has an advantage over artificial neural network (ANN) methods such as backpropagation [78] in that the training proceeds much faster [8] The linear programming in the current version of MSM T is implemented using the MINOS numerical optimization package [64] 2.4.2 ....

J. R. Quinlan. Decision trees as probabilistic classifiers. In Proceedings of Fourth International Workshop on Machine Learning, Los Altos, CA, 1987. Morgan Kaufmann.

....in this work and it has turned out that very simple pre pruning criteria suffice to contain this excessiveness. 2 As probabilistic classification has inspired this research, we would like to give credit, for some of the ideas that come up in this paper, to the work carried out by Quinlan [16]. However, we emphasize that the proposed approach does not relate to splitting using feature sets, as employed by C4.5 [18] Option decision trees as they were modified by Kohavi and Clayton [10] are quite close to our work. Kohavi and Clayton use a single structure for voting that is easier to ....

Quinlan, J.R., Decision trees as probabilistic classifiers. In Proceedings of the 4 th International Workshop on Machine Learning, Irvine, CA, 31-37, 1987.

.... diagnosis, economics, among others [16, 22, 24, 36] Several fuzzy learning algorithms for inducing rules from 4 given sets of data have been designed and used to good effect with specific domains [5 6, 8, 11, 15, 17 21, 30, 32 33] Strategies based on decision trees [9] were proposed in [10, 26 27, 30, 34 35], and Wang et al. proposed a fuzzy version space learning strategy for managing vague information [32] This paper integrates fuzzy set concepts with the apriori mining algorithm [4] and uses the result to find interesting itemsets and fuzzy association rules in transaction data with quantitative ....

J. R. Quinlan, "Decision tree as probabilistic classifier," The Fourth International Machine Learning Workshop, Morgan Kaufmann, San Mateo, CA, 1987, pp. 3137.

....tree algorithms are not able to handle the uncertainty in classification problems. Hence, their results are categorical and do not convey the uncertainty that may occur in the attribute values or in the case class. To overcome this limitation, Quinlan has developed probabilistic decision trees [5] where his major objective is to deal with examples characterized by missing or imprecise attribute values. However within his framework, only statistical uncertainty induced by information arisen from random behavior, is taken into account. In this paper, we present a classification method based ....

J. R. Quinlan "Decision trees as probabilistic classifiers" Proceedings of the Fourth International Workshop on Machine Learning, pp 31-37, June 22-25, 1987.

....results from the uncertainty encountered in the data. This uncertainty can appear either in the construction or in the classification phase. Ignoring it can a#ect the e#ciency of the obtained results. In order to overcome this drawback, probabilistic decision trees have been developed by Quinlan [6]. This kind of trees presents small extensions over the standard one and its use remains limited since it only deals with statistical uncertainty induced by information arisen from random behavior. The objective of this paper is to develop what we call a belief decision tree, a classification ....

Quinlan, J. R.: Decision trees as probabilistic classifiers. Proceedings of the Fourth international Machine Learning (1987) 31--37

....a line of research in this work and it has turned out that very simple pre pruning criteria suffice to contain this excessiveness. As probabilistic classification has inspired this research, we would like to give credit, for some of the ideas that come up in this paper, to the work carried out by Quinlan (1987). However, we emphasize that the proposed approach does not relate to splitting using feature sets, as employed by C4.5 (Quinlan, 1993) An earlier attempt to capitalize on these ideas appeared in (Kalles, 1994) Probably due to the immaturity of those results, the potential has not been fully ....

J.R. Quinlan. Decision trees as probabilistic classifiers. In Proceedings of the 4th International Workshop on Machine Learning, pages 31-37, Irvine, CA, June 1987.

....to learning by examples in machine learning (Kodratoff 1990, Michalski 1986, Michalski 1983) We compare here with two well known 16 Computational Intelligence methods. The decision tree approach (ID3) Quinlan 1983) generates a decision procedure instead of a rule. It was extended later (C4) (Quinlan 1987) to handle noisy data by associating each leaf with an error probability similar to the error probability attached to each conjunct in our work. No error probability for the entire decision tree is estimated even though it could be added with an analysis similar to ours. It does not address the ....

Quinlan, J.R. 1987. Decision trees as probabilistic classifiers. In Proc. 4th International Workshop on Machine Learning, Irvine, CA, 31--37.

....71 Learning with Missing Attribute Values: Every learning algorithm should handle missing attribute values in some way, because most of the real world datasets contain unknown attribute values. Therefore, in the literature, there are some methods to handle these kinds of attribute values [23, 43, 44, 46]. Most of the methods are based on one of the following ideas: ffl Ignoring examples that have unknown attribute value. ffl Assuming an additional special value for unknown attribute values. ffl Using probability theory by utilizing information provided by context. ffl Generating additional ....

....dimensions are independent from each other, no specialization is required. The concept descriptions can be overlapped. Another important property of the COFI algorithm is its way of handling the unknown attribute values. Most of the systems use ad hoc methods to handle the unknown attribute values [23, 44]. Like CFP, the COFI algorithm also ignores the unknown attribute values. Since the value of each attribute is handled separately, this causes no problem. The behavior of the COFI algorithm to the irrelevant features is very interesting. Irrelevant attributes can easily be detected by looking at ....

J.R. Quinlan, Decision Trees as Probabilistic Classifiers, In Proceedings of Fourth International Workshop on Machine Learning, pp: 31-37, June 1987.

....reliance on a specific, possibly unreliable, part of the system, allowing noisy effects to be over ridden and smoothed out by other model components. Statistical methods such as Bayesian techniques (e.g. 22] can be employed. AQ15 [19] performs weighted rule application in this way, and Quinlan [23] suggests how decision tree application can be made less brittle by introducing a degree of corroboration between decision tree branches. ffl The exception based paradigm For any system learning incrementally, noise presents the problem of when to ignore observations conflicting theory and ....

J. R. Quinlan. Decision trees as probabilistic classifiers. In P. Langley, editor, Proc. 4th International Workshop on Machine Learning, Kaufmann, Ca, 1987.

....by LMDT across a variety of classification tasks. To understand under what circumstances the bias of a multivariate tree (and LMDT s search bias for finding such a tree) is more appropriate than the bias of a univariate decision tree we compare LMDT to a univariate decision tree algorithm, C4.5 (Quinlan, 1987), across these tasks. The results of this comparison show that each approach has a selective superiority; for some of the tasks LMDT finds significantly more accurate trees than C4.5 and for others the reverse is true. Because the hypothesis space searched by a multivariate decision tree algorithm ....

Quinlan, J. R. (1987). Decision trees as probabilistic classifiers. Proceedings of the Fourth International Workshop on Machine Learning (pp. 31-37).

....threshold cause removing partitions more aggressively. If confidence threshold is zero then percentage of the noise in the concept description is equal to the noise level of the training set. Unknown Attribute Values Most of the real world data sets contain missing attribute values. Many authors [15, 27, 28, 29] were presented methods for handling unknown attribute values. Most of the methods are based on the following ideas: 1. Ignoring examples which have unknown attribute value. 2. Assuming an additional special value for unknown attribute values. This can lead to an anomalous situation [27] 3. ....

....anomalous situation [27] 3. Using probability theory by utilizing information provided by context. 4. Generating additional instances for all possible values of the unknown attribute [15] 5. Exploring all branches (on decision trees) remembering that some branches are more probable than others [28]. Although these methods for handling unknown attribute values sound promising on paper, they give unconvincing results. However, CFP handles unknown attributes very naturally, since it learns feature by feature in the case of an unknown attribute value it simply ignores processing of that ....

[Article contains additional citation context not shown here]

J. R. Quinlan. Decision Trees as Probabilistic Classifiers. In Proceedings of Fourth International Workshop on Machine Learning, pages 31--37, June 1987.

....sampling must be cheap to build and to use. At each iteration a new classifier is built (fortunately from a small sample) and then applied (unfortunately to a large sample) Our uncertainty sampling method also requires an estimate of the certainty of classifications (a class probability value) [28]; not all induction systems provide this. This paper examines a heterogeneous approach in which a classifier of one type selects instances for training a classifier of another type. It is motivated by applications requiring a type of classifier that would be too computationally expensive to use to ....

J.R. Quinlan. Decision trees as probabilistic classifiers. In Proceedings of the Fourth International Workshop on Machine Learning, pages 31--37, Irvine, California, 1987.

....noise. 5.2.3.3 Experiments with Increasing Level of Missing Values Most of the real world datasets contain missing (unknown) feature values and the percentage of missing values are shown in Table A.1. In order to cope with instances that contain missing values, several methods have been proposed [30, 55, 56, 57, 58]. These methods can be summarized as: ffl Ignoring instances which have unknown feature values. ffl Assuming an additional special value for unknown attribute values. ffl Using probability theory by utilizing information provided by context. ffl Generating additional instances for all possible ....

J.R. Quinlan, Decision Trees as Probabilistic Classifiers, In Proceedings of Fourth International Workshop on Machine Learning, 31--37, June 1987.

....j instance set j 2 j feature set j AND codelength(T) codelength(instances) THEN select (T) This rule checks to see if the initial choice of a univariate test was appropriate. The information theoretic measure does not provide reliable results if the number of training instances is too small (Quinlan, 1987b; Aha, 1990). In this cases, an instance based classifier is more likely to be appropriate because there is no minimum number of instances required to form an IBC. If the number of bits required to represent the univariate test and its corresponding error vector is more than the number of bits required to ....

Quinlan, J. R. (1987a). Decision trees as probabilistic classifiers. Proceedings of the Fourth International Workshop on Machine Learning (pp. 31-37). Irvine, CA: Morgan Kaufmann.

....correctness in a classification task, or using background knowledge which is not known with absolute certainty. Fortunately the systems for concept learning from examples (described in Section 3. 1) can be extended to produce some measure of probability in classification (e.g. AQ15 [27] and in ID3 [40]) However when learning is to occur in the presence of uncertain background knowledge, the problems become more difficult. Surprisingly (surprisingly because most AI applications use some method for representing uncertainty) little research into learning when uncertain background knowledge is ....

Quinlan, J. R. (1987). Decision trees as probabilistic classifiers. In Proc. 4th International Workshop on Machine Learning Ca, P. Langley, Ed., Kaufmann.

....thresholds, the classification result is very sensitive to noise in feature measurements. To minimise the sensitivity of tree classifiers to noise, several researchers have explored the use of soft decisions based on probabilities associated to the outcomes of the comparisons in the decision nodes [1, 5, 6]. Another approach is to transform decision trees into equivalent neural networks [2, 3, 7, 8, 11] Ivanova and N N N Y Y Y Y N Y N c1 a2 0.3 c2 c2 c3 c3 c2 a1 0.2 a3 0.4 a2 0.8 a1 0.6 Figure 1: A sample decision tree induced by ID3. It has a depth of 4 and contains 5 decision nodes. ....

J. R. Quinlan, Decision trees as probabilistic classifiers, In Proc. 4th International Workshop on Machine Learning, 31--37, 1987.

....of prototypicality ) of a concept. In AQ15, the degree to which the base concept representation of rules can be mixed with matching procedures can be varied [25] Quinlan recently proposed an adaptation of the decision tree interpreter to introduce a measure of probability of class membership [34]. A final point is that using matching processes can act as a useful source of domain knowledge, and can be used to resolve ambiguities in data. If a system is able to recognise near matches (i.e. near misses [49] as well as when an example matches an exemplar generalisation perfectly, then ....

J. R. Quinlan. Decision trees as probabilistic classifiers. In P. Langley, editor, Proc. 4th International Workshop on Machine Learning, Kaufmann, Ca, 1987.

No context found.

Quinlan, J.R. (1987a), Decision trees as probabilistic classifiers, in Langley(Ed), Proceedings of the Fourth International Workshop on Machine Learning, Los Altos: Morgan Kaufmann.

No context found.

Quinlan, J. R. (1987). Decision trees as probabilistic classifiers. Proceedings of the Fourth International Workshop on Machine Learning (pp. 31-37). Irvine, CA: Morgan Kaufmann.

No context found.

Quinlan, J. R. (1985), Decision Trees as Probabilistic Classifiers, School of Computing Sciences, New South Wales, Sydney, N.G.S., Australia.

No context found.

J. R. Quinlan. Decision Trees as Probabilistic Classifiers. In Fourth International Workshop on Machine Learning. Morgan Kaufmann, 1987.

No context found.

J. R. Quinlan, "Decision trees as probabilistic classifiers," in Proc. 4th Int. Workshop Machine Learning, Irvine, CA, 1987, pp. 31--37.

No context found.

J. R. Quinlan, Decision trees as probabilistic classifiers, Proceedings of the Fourth International Workshop on Machine Learning, pp 31-37, June 22-25, 1987.

First 50 documents

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