J. Ross Quinlan. Induction of decision trees. Machine Learning, 1:81--106, 1986.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....taken into account while building the tree. Fortunately, the problem of building a decision tree that has the above characteristics has been extensively studied by the machine learning community (under the name of tree induction) and a number of different heuristic algorithms have been developed [1, 29]. For our problem, we decided to use the C4.5 algorithm [30] as it leads to small trees, is computationally efficient, and as part of our earlier work, have developed efficient and scalable parallel formulations for it [14] Given a set of points A, each belonging to one of the k partitions, the ....

J. Ross Quinlan. Induction of decision trees. Machine Learning, 1:81--106, 1986.

....p ij log 2 (p ij ) H C = p i. log 2 (p i . HA = p .j log 2 (p . j) The information gain is the di#erence between the information stored in the cells and the information about class, that is to say: H T = H C HA This is the heuristic that was originally used by Quinlan in ID3 [Qui86] However, it has one significant drawback: it does not take into account the number of regions. If information gain were to be used raw without regard to the number of regions, then this would lead to a bias to having a huge number of regions. Imagine a region for each point in the space, such ....

J. Ross Quinlan. Induction of decision trees. Machine Learning, 1:81--106, 1986.

....## # # # # # # # # # # # ## Fig. 4. An example of decision tree that classifies documents. recipe #home #top) #recipe # pepper #pan) Fig. 5. An example of Boolean expression converted from the tree in Figure 4 We make the initial decision tree using an information gain measure[11] for greedy search without using any pruning technique. In our real case, the number of attributes (keywords) is large enough (several thousands) to make a tree that can correctly classify all examples in the training set training . Then for each path in the induced tree that ends in a ....

J. Ross Quinlan, "Induction of decision trees," Machine Learning, vol. 1, pp. 81--106, 1986.

....power supply [13] Machine Learning techniques, inside the supervised learning field, are automated procedures based on logical operations that learn a task starting from a suite of examples. In the classification field the attention has been centred, concretely, in approaches with decision trees [14], where classification is the result of a series of logical steps. These approaches are able to represent the most complex problems if they have enough data. Applied to the diagnosis, we can find these methods used for the classification of temporary patterns [15] or in previous works to the ....

J. Ross Quinlan. Induction of decision trees. Machine learning, 1986

....as we shall see throughout the paper. Although the classi er obtained is unlikely to be a perfectly linear classi er, any learning algorithm that is biased towards linearity is likely to learn better after the GCT is applied to the original data. The perceptron [45, 37] and decision trees [40, 41, 5] are just two examples of such classi ers. The perceptron is essentially a linear separating hyperplane of the form 0 i x i = 0, for which the i s are learned using a speci c iterative learning rule [49] based on gradient descent. Linear separating hyperplane classi ers for which the ....

J. Ross Quinlan. Induction of decision trees. Machine Learning, 1(1):81-106, 1986.

....of that applied by an person when features are added or removed, and that applied in deciding in what ways the learning algorithm should be restricted. Implicit background knowledge is easier to understand if we consider an example. There is a classic tennis playing problem, proposed by Quinlan [Qui86] and discussed in depth by Mitchell [Mit97] This problem involves determining if a given person will play tennis based on the weather. The observations are those of the weather, and belong to the class of all possible observations of weather, the example space. In the problem, the weather is ....

J. Ross Quinlan. Induction of decision trees. Machine Learning, 1:81--106, 1986.

....can be calculated. If the cut is good, these entropies will be smaller than the original entropy of the whole space. The difference of the average of the entropies of the two half spaces and the original entropy is the gain. We use information gain as defined in the classical ID3 algorithm [9]. For a given fitness function f from the domain the information gain of a concept C is defined as follows: ICI ) IS1 where function E is the entropy defined by E(p) p lnp (1 p) ln(1 p) Here p is the proportion of a given concept over the space under consideration. The natural ....

J. Ross Quinlan. Induction of decision trees. Machine Learning, 1(1):81-106, 1986.

....Second, the domain model is automatically modified as shown in Figure 1.15. A new concept orercial Seaport is added to the domain model as a subconcept of the original seaport. liarbor gert.Itarboar will map now into orercial eaport. Third, we apply machine learning algorithms (currently ID3 [Quinlan, 1986]) in order to obtain a concise description of this new concept. For example, it might construct a description that distinguishes commercial seaports from generic seaports by the number of cranes available. With this refined model, a query like retrieve all the seaports that have more than 15 ....

J. Ross Quinlan. Induction of decision trees. Machine Learning, 1(1):81-106, 1986.

....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, ....

J. Ross Quinlan (1986). Induction of Decision Trees. Machine Learning 1. 81106.

.... qt w DuA q q q h Az q c D z q c A D A q q q q q z q q q Az c Figure 6: A decision tree induced from web documents not home , and does not top belongs to class r . We make the initial decision tree using an information gain measure [Quinlan, 1986] for greedy search without using any pruning technique. In our real case, the number of attributes (keywords) is large enough (several thousands) to make a tree that can correctly classify all examples in the training set cbJdHegfheifkj . Then for each path in the induced tree that ends in a ....

J. Ross Quinlan. Induction of decision trees. Machine Learning, 1:81--106, 1986.

....etc. Neither the partitioning of the input space, nor the definition of the surfaces is known in advance. Also, the learning procedure is not biased towards simple surfaces either. A learning algorithm that has the capability to decompose the input space, such as CART [Breiman et al. 1984] or ID3 [Quinlan, 1986], may easily induce a solution of low error on the training set. However the solution may be an intricate function approximator that uses many decision boundaries in the input space. This is not a natural solution and may not prove to be general 28 either. A natural solution would have a short ....

....of rules. In our system no special seeding is necessary, and more importantly, we do not provide any help on how the problem can be decomposed. Classical machine learning approaches to decomposition using the divide and conquer paradigm are techniques such as CART [Breiman et al. 1984] and ID3 [Quinlan, 1986] (or its successor C4.5) Our approach appears suitable for online learning and adaptation to changing environments. The adaptive mixture of experts (ME) Jacobs et al. 1991] is a modular neural network approach which is the closest in spirit to our problem decomposition approach. ME starts with ....

J. Ross Quinlan, "Induction of Decision Trees," Machine Learning, pages 81--106, 1986.

....the first phase of the search this function is the associated probability of a rule, and in the second phase it is the cover of the rule (while retaining the maximal probability found) 10 This problem is well documented in the machine learning literature. Older solutions are ID3, AQ15 and CN2 [9, 6, 2]. More modern, non deterministic approaches use genetic algorithms or simulated annealing. See [4] for an overview of a some of these systems. For our problem, a non deterministic approach is best suited. For, there can be many different (100 Gamma ffi) homogeneous discoveries, that describe ....

J. Ross Quinlan. Induction of decision trees. Machine Learning, 1:81 -- 106, 1986.

....The resulting model is not necessarily a qualitative one. Several machine learning techniques have been proposed so far, including the induction of deci AI Communications ISSN 0921 7126, IOS Press. All rights reserved 2 Wotawa et al. Deriving Qualitative Rules from Neural Networks sion trees [17], association rules [13] or neural networks [7] Bratko, Muggleton, and Varsek [1] introduced another interesting approach for deriving qualitative models out of available data. Their work is quite similar to our work with one exception. They use inductive logic programming techniques for ....

....in cloud cover reduces the photochemically active radiation and thus ozone production. 5. Other Approaches In this section we compare the outcome of our methods with the outcome of two machine learning and data mining methods applied to the same data sets. The first method is the ID3 algorithm [17] computing a decision tree from data. The second method computes a set of association rules from data using frequent sets [13] For the evaluation we convert the available numerical data from the ozone forecasting domain to a qualitative data set using the same mapping as for our neural network ....

[Article contains additional citation context not shown here]

J. Ross Quinlan. Induction of decision trees. Machine Learning, 1(1):81--106, 1986.

No context found.

J. Ross Quinlan. Induction of decision trees. Machine Learning, 1:81--106, 1986.

No context found.

John R. Quinlan. 1986. Induction of decision trees. Machine Learning, 1(1):81--106.

No context found.

J. Ross Quinlan. Induction of decision trees. Machine Learning, 1(1):81--106, 1986.

No context found.

J. Ross Quinlan. Induction of decision trees. Machine Learning, 1:81--106, 1986.

No context found.

J. Ross Quinlan. Induction of decision trees. Machine Learning, 1:81--106, 1986.

No context found.

J. Ross Quinlan. Induction of decision trees. Machine Learning, 1:81--106, 1986.

No context found.

J. Ross Quinlan, Induction of Decision Trees, Machine Learning 1(1), 1986, pp. 81--106.

No context found.

J. Ross Quinlan. Induction of decision trees. Machine Learning, 1(1):81--106, 1986.

No context found.

J. Ross Quinlan, Induction of Decision Trees, Machine Learning 1(1), 1986, pp. 81--106.

No context found.

J. Ross Quinlan. Induction of decision trees. Machine Learning, 1:81--106, 1986.

No context found.

J. Ross Quinlan. 1986. Induction of decision trees. Machine Learning, 1:81-106.

No context found.

John Ross Quinlan. Induction of decision trees. Machine Learning, 1:81#106, 1986.

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