A. K. Spackman, Learning Categorical Decision Criteria in Biomedical Domains, In Proceedings of the Fifth International Conference on Machine Learning, University of Michigan, Ann Arbor, 1988.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....An additional advantage is the easy and natural handling of missing feature values. The rationale behind this knowledge representation is that humans maintain knowledge in this form, especially in medical domains. An example for this approach is presented, called CRiteria Learning System [66]. It aims to learn decision rules in the form of criteria tables as humans do. One of the shortcomings of feature projections is that descriptions involving a conjunction between two or more features can not be represented. This chapter discusses the CFP, COFI, and k NNFP algorithms that use ....

A. K. Spackman, Learning Categorical Decision Criteria in Biomedical Domains, In Proceedings of the Fifth International Conference on Machine Learning, University of Michigan, Ann Arbor, 1988.

....is maintained as the projections of the training set on each feature dimension separately. The rationale behind this knowledge representation is that humans maintain knowledge in this form, especially in medical domains. An example for this approach is the CRiteria Learning System (CRLS) [72], which aims to learn decision rules in the form of criteria tables as humans do. The most important advantage of this representation is that the projections of the feature values can be organized for each feature in a way that it reduces the time for the computation of similarity to all training ....

A. K. Spackman, Learning Categorical Decision Criteria in Biomedical Domains, In Proceedings of the Fifth International Conference on Machine Learning, University of Michigan, Ann Arbor, 1988.

.... learning algorithms were actually applied, for example, two classification algorithms are used in localization of primary tumor, prognostics of recurrence of breast cancer, diagnosis of thyroid diseases, and rheumatology [4] Another example is the CRLS system applied to a biomedical domain [5]. This paper presents a new machine learning algorithm for another medical problem, which is the diagnosis of cardiac arrhythmia from standard 12 lead ECG recordings. The algorithm is called VFI5 for Voting Feature Intervals. The VFI5 algorithm is similar to the VFI algorithm [2] which has been ....

Spackman A. K. Learning Categorical Decision Criteria in Biomedical Domains, In Proceedings of the Fifth International Conference on Machine Learning, University

.... actually applied; for example, two classi cation systems are used in the localization of primary tumor, the prognostics of recurrence of breast cancer, the diagnosis of thyroid diseases, and in rheumatology [12] The CRLS is a system for learning categorical decision criteria in biomedical domains [17]. VFI5, a feature projection based learning system, was successfully applied to di erential diagnosis of erythemato squamous diseases [9] Classi cation learning algorithms are composed of two components; namely, the training and the prediction (classi cation) The training phase, using some ....

A. K. Spackman, \Learning categorical decision criteria in biomedical domains", Proc. of the Fifth ICML, 1988, pp. 36-46.

....specializes the hypothesis by adding an attribute value pair and increasing M by one. With these two operators, Generate M of N has the potential of constructing any M of N attribute that can be defined by a set of primitive attributes. It is worth mentioning that learning systems such as Crls [Spackman, 1988] and MoN [Ting, 1994] also construct M of N attributes, but they represent learned theories in the form of production rules (called M of N rules) rather than decision trees. Crls generates a set of M of N rules directly from training data, while MoN constructs M of N rules 49 based on production ....

....production rules (called M of N rules) rather than decision trees. Crls generates a set of M of N rules directly from training data, while MoN constructs M of N rules 49 based on production rules generated by C4.5rules. Crls is shown to outperform standard rule learning in several medical domains [Spackman, 1988]. Ting [1991] demonstrates the performance advantage of MoN as well as ID2 of 3 over C4.5rules in terms of higher prediction accuracy and smaller theory size in a biology domain (Splice junction) Neither, a propositional theory refinement system [Baffes and Mooney, 1993] is capable of revising ....

[Article contains additional citation context not shown here]

K.A. Spackman, Learning categorical decision criteria in biomedical domains. Proceedings of the Fifth International Conference on Machine Learning, San Mateo, CA: Morgan Kaufmann, 36-46.

....restrictive property) the problem cannot in general be decomposed this way, and the observed values of f(x) for unobserved x are in general restricted. In [2] we consider some other possible restrictions on f (when it is a boolean function) More examples for similar problems can be found in [3, 14, 15]. In this paper we show that a network flow model can be used to determine the best approximation f when the partial ordering of the outcomes is an interval order. An interval order is one in which when every element can be associated with an interval of real numbers, such that ff fi if and only ....

Spackman, K.A., "Learning Categorical Decision Criteria in Biomedical Domains", Proceedings of the Fifth International Conference on Machine Learning, (1988), 36-46.

....two of which are most relevant to the enterprise of establishing method task fits. They observe that the type of classifier needed for some domains favors a network approach. For example, m out of n descriptions of classes look for at least m of a possible n conditions to be satisfied; Spackman (1988) notes that such descriptions are common in biomedical domains. While m out of n descriptions can be represented naturally in a network, their propositional form requires one disjunct for every way of selecting a subset of size m, making 3 Domain Network Decision Tree size atts classes error ....

Spackman, K.A. Learning categorical decision criteria in biomedical domains. Proceedings of the Fifth International Conference on Machine Learning, Ann Arbor, June 1988.

.... process working with some sort of relational description language, e.g. 28, 11,9] In several recent cases, this type of approach has focussed on what are known as counting or M of N features, i.e. features which effectively count the number of occurrences of a particular variable value, cf. [29,30,2,8 31,32]. Equating constructive induction with relational learning is also compatible with the intuition (noted above) that constructive induction involves the creation of non local partitions. Any feature function whose values depend on absolutes will tend to define a partitioning involving contiguous ....

Spackman, K. (1988). Learning categorical decision criteria in biomedical domains. Proceedings of the Fifth International Conference on Machine Learning (pp. 36-46). San Mateo, CA: Morgan Kaufmann,.

.... process working with some sort of relational description language, e.g. 29, 11,10] In several recent cases, this type of approach has focussed on what are known as counting or M of N features, i.e. features which effectively count the number of occurrences of a particular variable value, cf. [30,31,2,9 32,33]. It is worth noting, finally, that although all these approaches place the constructive process within the domain of supervised methods, there is no reason why constructive induction cannot form a part of an unsupervised process. The justification analysis shows that it can form a part of any ....

Spackman, K. (1988). Learning categorical decision criteria in biomedical domains. Proceedings of the Fifth International Conference on Machine Learning (pp. 36-46). San Mateo, CA: Morgan Kaufmann,.

....interested in creating composite m of n features for several reasons. First, m of n concepts more closely resemble fuzzy categories with graded structure (Barsalou, 1985; Smith Medin, 1981) Second, there is some evidence that this bias helps in the acquisition of naturally occurring concepts (Spackman, 1988). For example, a successful medical expert system makes use of criteria tables that are essentially m of n concepts (Kingsland, 1985) Our motivation is somewhat similar to that of Utgoff (1988) in developing perceptron trees. In particular, the terms constructed to serve as tests at nodes in ....

Spackman, K. (1988). Learning categorical decision criteria in biomedical domains. Proceedings of the Fifth International Workshop on Machine Learning (pp. 36--46). Ithaca, NY: Morgan Kaufmann.

.... 0 6 19 2 79.4 Liver disorders 345 0 0 6 6 2 58.0 Diabetes 768 0 0 8 8 2 65.1 Wisconsin breast cancer 699 0 0 9 9 2 65.5 Promoters 106 0 57 0 57 2 50.0 Nettalk phoneme 5438 0 7 0 7 52 18.7 Nettalk stress 5438 0 7 0 7 5 40.1 Nettalk letter 5438 0 7 0 7 163 11.2 Tic tac toe 958 0 9 0 9 2 65.3 domains (Spackman, 1988). The objective of using a test suite with this property is to test whether the algorithms capable of learning this kind of concept can work well in some real world applications. In all the experiments reported throughout this paper, all the algorithms are run with their default option settings. ....

....concepts that are often studied in the machine learning community belong to these types of domain. Conjunctions and disjunctions are often used by humans to represent knowledge and appear in many real world domains. M of N like concepts are used in some real world domains such as medical domains (Spackman, 1988). Although a single nominal X of N representation can represent more complex 20 ZIJIAN ZHENG Table 8. Results of C4.5 and XofN. It is shown that XofN can significantly improve the performance of decision tree learning, but it suffers from the fragmentation problem in the DNF, CNF, and ....

[Article contains additional citation context not shown here]

Spackman, K. (1988). Learning categorical decision criteria in biomedical domains. Proceedings of the Fifth International Conference on Machine Learning (pp. 36--46). San Mateo, CA: Morgan Kaufmann.

....represented using some form of partial matching or evidence summing, such as M of N concepts, which are true if at least M of a set of N specified features are present in an example. There has been some work on the induction of M of N rules that demonstrates the advantages of this representation [ Spackman, 1988; Murphy and Pazzani, 1991 ] Other work has focused on revising rules that have real valued weights [ Towell and Shavlik, 1992; Mahoney and Mooney, 1992 ] However, revising theories with simple M of N rules has not previously been addressed. Since M of N rules are more constrained than This ....

....80.00 85.00 90.00 95.00 100.00 105.00 110.00 115.00 120.00 125.00 130.00 135.00 140.00 145.00 150.00 0.00 20.00 40.00 60.00 80.00 Figure 5: Concept Complexity the same as Neither MofN s. 4 Related Work Several researchers have developed methods for inducing M of N concepts from scratch. CRLS [ Spackman, 1988 ] learns M of N rules and out performed standard rule induction in several medical domains. ID 2 of 3 [ Murphy and Pazzani, 1991 ] incorporates M of N tests in decisiontree learning and out performed standard decision tree induction in a number of domains. Both projects clearly demonstrate the ....

K. A. Spackman. Learning categorical decision criteria in biomedical domains. In Proceedings of the Fifth International Conference on Machine Learning, pages 36--46, Ann Arbor, MI, June 1988.

.... a construction of descriptions that involve counting properties (e.g. that M properties out of N possible properties are present in an object) which may be additionally combined with logical conditions (e.g. in the DNF form) Problems of this type occur in many real world problems (e.g. Spackman, 1988; Towell Shavlik, 1994) The proposed solution is based on the application of a new type of constructive induction rule, counting attribute generation rule, which explores an attribute symmetry in generated hypotheses. Such a symmetry is indicated by the presence of the exclusive or or ....

.... exactly two of the six attributes have their first value, which is a special case of the M ofN concept. There have been several efforts concerned with learning Mof N concepts. For example, the system CRLS learns Mof N rules by employing non equivalence symmetry bias and criteria tables (Spackman, 1988), ID 2 of 3 incorporates M of N tests in decision trees (Murphy Pazzani, 1991) AQ17 DCI (Bloedorn Michalski, 1991) employs a variety of operators to construct new attributes, NEITHER MofN (Baffes Mooney, 1993) is able to refine M of N rules by increasing or decreasing either of M or N. ....

Spackman, K.A., "Learning Categorical Decision Criteria in Biomedical Domains," Proc. of the 5th International Conference on Machine Learning, Morgan Kaufmann, San Mateo, CA, pp. 36-46, 1988.

....is maintained as the projections of the training set on each feature dimension separately. The rationale behind this knowledge representation is that humans maintain knowledge in this form, especially in medical domains. An example for this approach is the CRiteria Learning System (CRLS) [72], which aims to learn decision rules in the form of criteria tables as humans do. The most important advantage of this representation is that the projections of the feature values can be organized for each feature in a way that it reduces the time for the computation of similarity to all training ....

A. K. Spackman, Learning Categorical Decision Criteria in Biomedical Domains, In Proceedings of the Fifth International Conference on Machine Learning, University of Michigan, Ann Arbor, 1988.

....concept, experiments are repeated ten times using different training and test sets. In addition to the fourteen artificial domains, ten real world domains from the UCI repository of machine learning databases [Murphy and Aha, 1996] are used, in which M of N like concepts are expected to be found [Spackman, 1988]. They consist of five medical domains (Cleveland Heart Disease, Hepatitis, Liver Disorders, Pima Indians Diabetes, Wisconsin Breast Cancer) one molecular biology domain (Promoters) three linguistics domains (Nettalk(Phoneme) Nettalk(Stress) Nettalk(Letter) and one game domain (Tic Tac Toe) ....

K.A. Spackman, Learning categorical decision criteria in biomedical domains. Proceedings of the Fifth International Conference on Machine Learning, San Mateo, CA: Morgan Kaufmann, 36-46.

....A simple example of a conditional counting rule is the M of N concept. Here, the counting condition is at least M out of N properties of some kind must present in an object, and the logic type condition is null. Problems of this type occur in many real world problems, for example, in medicine (Spackman, 1988), planning (Callan Utgoff, 1991) game playing (Fawcett Utgoff, 1991) biology (Baffes Mooney, 1993) and biochemistry (Towell Shavlik, 1994) The proposed method can learn DNF descriptions (decision rules) one or more counting conditions (e.g. M of N concepts) or any combination of the ....

....a learning system needs to be able to form what we call a counting condition. There have been several early efforts in this research direction. For example, the CRLS system learns M of N rules (a special case of a counting condition) by employing non equivalence symmetry bias and criteria tables (Spackman, 1988). The ID 2 of 3 system incorporates M of N tests in decision trees (Murphy Pazzani, 1991) The AQ17 DCI program employs a variety of datadriven operators to construct new attributes (Bloedorn Michalski, 1991) Fawcett and Utgoff (1991) used feature representation similar to the Michalski s ....

Spackman, K.A. (1988). Learning Categorical Decision Criteria in Biomedical Domains. In Proceedings. of the 5th International Conference on Machine Learning (pp. 36-46). San Mateo, CA: Morgan Kaufmann.

....to ensembles whose members are more syntactically diverse. We chose a variety of domains for this study: LED 8 and KRK 100 are noise free, Diabetes and Iris may contain class and attribute noise and the Splice domain may contain classes which can be succintly described with m of n rules (e.g. Spackman, 1988). Table 9 shows the accuracies obtained by combining eleven stochastically generated models using the Uniform Voting evidence combination method. Our hope is that increasing the bucket size will lead to an increase in ensemble accuracy. However, as Table 9 shows, increasing the bucket size does ....

Spackman, K. (1988.) Learning Categorical Decision Criteria in Biomedical Domains In Proceedings of the Fifth International Conference on Machine Learning Ann Arbor, MI: Morgan Kaufmann.

....a consistent hypothesis with an error rate less than 10 on any test set (see [Pagallo, 1990] for details) On each concept, experiments are repeated ten times using different training and test sets. In addition, we use ten real world domains on which M of N like concepts are expected to be found [Spackman, 1988]. They are five medical domains: Cleveland Heart Disease, Hepatitis, Liver Disorders, Pima Indians Diabetes, Wisconsin Breast Cancer, one molecular biology domain: Promoters, three linguistics domains: Nettalk(Phoneme) Nettalk(Stress) Nettalk(Letter) and one game domain: Tic Tac Toe. ....

K.A. Spackman, Learning categorical decision criteria in biomedical domains. Proceedings of the Fifth International Conference on Machine Learning, 36-46, Morgan Kaufmann, 1988.

....test set (the whole universe) are given by the problem designers. In our experiments, we follow this methodology and run experiments once on the given training set and test set for each domain. In addition, ten real world domains are used, on which M of N like concepts are expected to be found [ Spackman, 1988 ] They are five medical domains (Cleveland Heart Disease, Hepatitis, Liver Disorders, Pima Indians Diabetes, and Wisconsin Breast Cancer) one molecular biology domain (Promoters) three linguistics domains (Nettalk(Phoneme) Nettalk(Stress) and Nettalk(Letter) and one game domain ....

....accuracy. Compared to ID2 of 3, XofN learns smaller trees on seven out of ten domains. 5 Discussion It has been found that XofN works quite well on a set of artificial and real world domains. We expect that XofN can be applied to domains containing M of N concepts, such as biomedical domains [ Spackman, 1988 ] linguistic domains, and domains containing parity concepts found for example in digital logic circuit design. To apply XofN to domains containing complex DNF concepts with long terms, some mechanisms are necessary to overcome the fragmentation problem. One approach is using subsetting of ....

K.A. Spackman, Learning categorical decision criteria in biomedical domains. Proceedings of the Fifth International Conference on Machine Learning, 36-46, Morgan Kaufmann, 1988.

....best represented using some form of partial matching or evidence summing, such as M of N concepts, which are true if at least M of a set of N specified features are present in an example. There has been some work on the induction of M of N rules demonstrating the advantages of this representation [17, 9]. Other work has focused on revising rules that have real valued weights [19, 6] However, revising theories with simple M of N rules has not previously been addressed. Since M of N rules are more constrained than rules with real valued weights, they provide a stronger bias and are easier to ....

....though Neither MofN was unable to exactly duplicate the original theory in all cases, the refinements made seemed reasonable in light of the alterations made in the modified theories. 4 Related Work Several researchers have developed methods for inducing M of N concepts from scratch. CRLS [17] learns M of N rules and out performed standard rule induction in several medical domains. ID 2 of 3 [9] incorporates M of N tests in decision tree learning and outperformed standard decision tree induction in a number of domains. Both projects clearly demonstrate the advantages of M of N rules. ....

K. A. Spackman. Learning categorical decision criteria in biomedical domains. In Proceedings of the Fifth International Conference on Machine Learning, pages 36--46, Ann Arbor, MI, June 1988.

.... were actually applied; for example, two classification systems are used in localization of primary tumor, prognostics of recurrence of breast cancer, diagnosis of thyroid diseases, and rheumatology [10] The CRLS system is a system for learning categorical decision criteria in biomedical domains [15]. The case based BOLERO system learns both plans and goals states, with the aim of improving the performance of a rule based system by adapting the rule based system behavior to the most recent information available about a patient [13] The DIAGAID is a program, using connectionist approach, to ....

A. K. Spackman, Learning Categorical Decision Criteria in Biomedical Domains, in: Proceedings of the Fifth International Conference on Machine Learning, (University of Michigan, Ann Arbor, 1988) 36-46. Figure and Table Captions

No context found.

Spackman, K. "Learning Categorical Decision Criteria in Biomedical Domains ", in Proceeding of the 5th International Workshop on Machine Learning, p36-46, 1988.

No context found.

Spackman, K.A., "Learning Categorical Decision Criteria in Biomedical Domains," Proceedings of the Fifth International Conference on Machine Learning, Morgan Kaufman, San Mateo, CA, pp. 36-46, 1988.

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