Zheng Z. Constructing New Attributes for Decision Tree Learning//In: A thesis submitted in fulfillment of the requirements for the degree of Doctor of Philosophy. The University of Sydney, Australia, 1996. $. HBaXHeHKO A.F., 3akIqeHKO IO.lq., [kIMkITpOB B.. lqpviltaTVie pememih Ha OCHOBe caMoopraHH3amm. MocIBa, COBeTCime paio, 1976- 280 CTp.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....new attributes as conjunctions, disjunctions or negations of basic attributes are useful. The Fringe method [6] for example, constructs new attributes directly from the branches of decision tree, using constraints from nodes. The CI method evaluates new attributes generated from decision tree [7]. Classification problems have relatively simple graphical interpretation. The task of classification is completed when a space of attributes is divided into regions and each record of data set is located within such region according to the class of data record. All decision trees methods that ....

Zheng Z. Constructing New Attributes for Decision Tree Learning//In: A thesis submitted in fulfillment of the requirements for the degree of Doctor of Philosophy. The University of Sydney, Australia, 1996. $. HBaXHeHKO A.F., 3akIqeHKO IO.lq., [kIMkITpOB B.. lqpviltaTVie pememih Ha OCHOBe caMoopraHH3amm. MocIBa, COBeTCime paio, 1976- 280 CTp.

....that are difficult to understand. Multivariate tests are more difficult to understand, but usually produce simpler, more compact trees. In this section we only review the work concerned with building linear combinations of ordered attributes. The work of Pagallo and Haussler [PH90] and Zheng [Zhe96] referred to earlier (section 1.3.2) is oriented towards building logical combinations of boolean attributes. Also, higher degree polynomials could be considered here. Nevertheless most authors prefer to consider only linear combinations mainly due to their simplicity. Breiman et al. BFOS84] ....

....is M of N including the variants at least Mof N, at most M of N, and exactly M of N. A M of N operator generates Boolean attributes. Given an integer M and a set of N conditions based on existing attributes, an at least M of N attribute is true if at least M of the N conditions are true. Zheng [Zhe96] proposes the X of N constructor that returns the number of true conditions. It generates ordered discrete values. One of the quite well known constructive induction system is the FRINGE family of algorithms [PH90] These algorithms iteratively build a decision tree based on the existing ....

Z. Zheng. Constructing New Attributes for Decision Tree Learning. PhD thesis, University of Sydney, 1996.

....M of N including the variants at least M of N, at most M of N, and exactly M of N. A M of N operator generates boolean attributes. It consists on a value M and a set of N conditions based on existing attributes. An at least M of N attribute is true if at least M of the N conditions are true. Zheng[18] proposes the X of N constructor that returns the number of true conditions. It generates ordered discrete values. One of the more divulged constructive induction system is the FRINGE family of algorithms [11] All algorithms from this family, iteratively build a decision tree based on the ....

Zheng, Z "Constructing New Attributes for Decision Tree Learning", Ph.D. Thesis, University of Sydney,

....We use experiments to evaluate the performance of the three CI algorithms by comparing them with C4.5 and SFRINGE in a set of artificial and natural domains. C4.5 is used as the baseline for this empirical study. SFRINGE is our implementation of the FRINGE algorithm (Pagallo, 1990) with extensions (Zheng, 1996). It follows the idea of SYMFRINGE (Yang et al., 1991) For each leaf, SFRINGE constructs one new attribute using the conjunction of two conditions at the parent and grandparent nodes of the leaf. Note that SFRINGE uses the same tree generation method, tree pruning method, tree evaluation function, ....

....CI3. Table 5 shows prediction accuracies and theory complexities for the CI algorithms with the R2strategy divided by the corresponding values for the CI algorithms with the L2 strategy in the set of artificial and natural domains. The detailed accuracies and theory complexities can be found in Zheng (1996). Constructing Conjunctive Attributes using Production Rules Journal of Research and Practice in Information Technology, Vol. 32, No. 1, February 2000 25 9 For the fixed path based approach, the number of new attributes of size two that could be created is equal to the number of decision nodes ....

[Article contains additional citation context not shown here]

ZHENG, Z. (1996): Constructing New Attributes for Decision Tree Learning, Ph.D. Thesis, Basser Department of Computer Science, The University of Sydney.

....learning algorithms, namely SFringe, CI3, CAT, and ID2 of 3. C4.5 is used as the baseline for the comparisons since most of these constructive induction algorithms use it as their selective induction component. SFringe is our implementation of the Fringe algorithm (Pagallo, 1990) with extensions (Zheng, 1996). It follows the idea of SymFringe (Yang, Rendell, Blix, 1991) For each leaf, SFringe constructs one new attribute using the conjunction of two conditions at the parent and grandparent nodes of the leaf. CI3 (Zheng, 1992, 1996) and CAT (Zheng, 1998) are also constructive decision tree learning ....

....implementation of the Fringe algorithm (Pagallo, 1990) with extensions (Zheng, 1996) It follows the idea of SymFringe (Yang, Rendell, Blix, 1991) For each leaf, SFringe constructs one new attribute using the conjunction of two conditions at the parent and grandparent nodes of the leaf. CI3 (Zheng, 1992, 1996) and CAT (Zheng, 1998) are also constructive decision tree learning algorithms. CI3 creates new attributes from production rules that are transformed from a decision tree. For each rule, it uses the conjunction of two conditions near the root of the tree as a new attribute (default option setting ....

[Article contains additional citation context not shown here]

Zheng, Z. (1996). Constructing New Attributes for Decision Tree Learning. Doctoral dissertation, Basser Department of Computer Science, The University of Sydney [available at http://www3.cm.deakin.edu.au/~zijian/Papers/thesis.ps.gz].

No context found.

Z. Zheng, Constructing New Attributes for Decision Tree Learning, Ph.D. Thesis, Basser Department of Computer Science, The University of Sydney (1996).

....from the raw tree, and are used for building decision trees in subsequent iterations. They may consist of both primitive and new attributes. Finally, CAT selects the best pruned tree from the two stages as its output. CAT uses an MDL inspired heuristic function as its tree evaluation function [7]. The function is similar to the coding cost function used by Quinlan and Rivest [8] but exceptions of a tree on the training set are replaced by the more pessimistically estimated exceptions of the tree [9] In addition, new attributes are encoded [7] In the current implementation, CAT adopts ....

.... function as its tree evaluation function [7] The function is similar to the coding cost function used by Quinlan and Rivest [8] but exceptions of a tree on the training set are replaced by the more pessimistically estimated exceptions of the tree [9] In addition, new attributes are encoded [7]. In the current implementation, CAT adopts the following stopping criterion: ffl No new attribute can be constructed, or ffl No better pruned tree has been built in five consecutive iterations, or ffl A given maximum iteration number is reached (the default value is 20) Note that five ....

[Article contains additional citation context not shown here]

Z. Zheng, Constructing New Attributes for Decision Tree Learning, Ph.D. Thesis, Basser Department of Computer Science, The University of Sydney, 1996.

....this section, to evaluate the performance of the three CI algorithms by comparing them with C4.5 and SFringe in a set of artificial and real world domains. C4.5 is used as the baseline for this empirical study. SFringe is our implementation of the Fringe algorithm (Pagallo, 1990) with extensions (Zheng, 1996, pp. 211ff) It follows the idea of SymFringe (Yang et al. 1991) For each leaf, SFringe constructs one new attribute using the conjunction of two conditions at the parent and grandparent nodes of the leaf. Note that SFringe uses the same tree generation method, tree pruning method, tree ....

....Table 5 shows prediction accuracies and theory complexities for the CI algorithms with the R2 strategy divided by the corresponding values for the CI algorithms with the L2 strategy in the set of artificial and real world domains. The detailed accuracies and theory complexities can be found in Zheng (1996, pp. 215 217) From Table 5, we can see that the average accuracy of CI1 with the R2 strategy across all these domains is the same as that of CI1 with the L2 strategy. The situation is the same for CI2. The average accuracy of CI3 with the R2 strategy is slightly lower than that of CI3 with the ....

[Article contains additional citation context not shown here]

Zheng, Z. (1996): Constructing New Attributes for Decision Tree Learning, Ph.D. Thesis, Basser Department of Computer Science, The University of Sydney.

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