J.D. Kelly and L. Davis. Hybridizing the genetic algorithm and the k nearest neighbors classification algorithm. In Proc. of the 4th International Conference on Genetic Algorithms and their Applications (ICGA'91), pages 377--383, 1991. 60  Home/Search   Document Not in Database   Summary   Related Articles   Check

....not participate. Each resulting subset of features is evaluated according to its classification accuracy on a set of testing data using a nearest neighbor classifier. This technique was later expanded to allow linear feature extraction, by Punch et al. 25] and independently by Kelly and Davis [26]. The single bit associated with each feature is expanded to a real valued coefficient, allowing independent linear scaling of each feature, while maintaining the ability to remove features from consideration by assigning a weight of zero. Given a set of feature vectors of the form X = xl,x2, ....

J. D. Kelly and L. Davis, "Hybridizing the genetic algorithm and the k nearest neighbors classification algorithm," in Proceedings of the Fourth International Conference on Genetic Algorithms and their Applications, 1991, pp. 377-383.

.... regions of the instance space similar to decision trees induced by a splitting algorithm (Rendell, 1983, 1985; Sikora Shaw, 1994) to expertsystem rules (Montana, 1990) to weights for a game s evaluation function (Rendell, 1990) to weights and orientations for the k nearest neighbor algorithm (Kelly Davis, 1991; Punch et al. 1993) to finite state automata (Fogel et al. 1966) and context free grammars (Wyard, 1991) to production system like rules (Booker et al. 1989; De Jong et al. 1993; Greene Smith, 1993, 1994; Holland, 1986; Janikow, 1993) Financial Forecasting Using Genetic Algorithms 549 ....

Kelly, J. D., Jr., and L. Davis. 1991. Hybridizing the genetic algorithm and the k-nearest neighbors classification algorithm. In Proceedings of the fourth international conference on genetic algorithms, 377383.

....of the supervising teacher. The ReMind case based reasoning development shell [Cognitive Systems, Inc. 1990] also incorporates a facility for the user to create prototypes to further index a case base. Genetic algorithm classification systems have been created by Vafaie and De Jong [1992] and by Kelly and Davis [1991] to select features by learning real valued weights for features in a data set and by DeJong and Spears [1991] to learn conceptual classification rules. Tan and Schlimmer [1990] have shown how features for nearest neighbor retrieval may be selected where the determination of feature values has a ....

Kelly, J. D. J. and Davis, L. 1991. Hybridizing the Genetic Algorithm and the K Nearest Neighbors Classification Algorithm. In Proceedings, Fourth International Conference on Genetic Algorithms, San Diego, CA. Morgan Kaufmann, San Mateo, CA. 377--383.

....selection using genetic algorithms. The exhaustive search would have meant prohibitive calculation of 2 400 combinations of features. Feature weights The k NN classifiers can be further enhanced by modifying the feature space, or equivalently, changing the weights in the distance measure (Kelly and Davis 1991). A commonly used weighted Euclidean metric between two vectors X and Y in an N dimensional feature space is defined as: d = i x i y ( 2 =1 N . 1 2 By changing the weights, i , the shape of the feature space can be changed. See Figure 1. The feature selection is a trivial case ....

Kelly, J. D., and L. Davis. 1991. Hybridizing the genetic algorithm and the k nearest neighbors classification algorithm. Fourth International Conference on Genetic Algorithms and their Applications. 377-83.

....to decision trees to speed up hillclimbing. In contrast, the caching scheme used by the programs presented here was a simple memoizing technique that relied on exact match of function arguments. Genetic algorithm classification systems have been created by Vafaie and De Jong [1992] and by Kelly and Davis [1991] to select features by learning real valued weights for features in a data set and by DeJong and Spears [1991] to learn conceptual classification rules. Tan and Schlimmer [1990] have shown how features for nearest neighbor retrieval may be selected where the determination of feature values has a ....

Kelly, J. D. J. and Davis, L. 1991. Hybridizing the Genetic Algorithm and the K Nearest Neighbors Classification Algorithm. In Proceedings, Fourth International Conferenceon Genetic Algorithms, San Diego, CA. Morgan Kaufmann, San Mateo, CA. 377--383.

....approach. In particular finding an appropriate distance measure, dealing satisfactorily with noise (that essentially corresponds to atypical problems in our case) and superfluous attributes (features) are the most prominent problems for which (partial) remedies already exist (e.g. AK89] [KD91], FA96] ....

Kelly, J.D.; Davis, L.: Hybridizing the genetic algorithm and the k nearestneighbor classification algorithm, Proc. 4 th International Conference on Genetic Algorithms, San Diego, CA, USA, 1991.

....or the plaintiff ) An unknown sample can be placed in the space based on its feature values and then can be classified based on its k nearest neighbors where k is set to some integer value. The distance between feature vectors is defined as d ij = f P n a=1 (Xia Gamma X ja) 2 g 1 2 ((Kelly Davis 1991) Where i and j are the i th and j th vectors of features being compared. Xia is the value of the a th attribute for the i th vector X ja is the value of the a th attribute for the j th vector. This gives a Euclidean measure of distance, even though the features are symbolic. Since ....

Kelly, J. & Davis, L. 1991. Hybridizing the Genetic Algorithm and the K Nearest Neighbors Classification Algorithm. Proceedings of the 4th International Conference on Genetic Algorithms and their Applications. Morgan Kaufmann:CA.

....which probably reflects that an increased separation between the classes gave better resolution. This is precisely the approach we have taken, using the GA to discover real valued weights that modify the Knn space. We pursued this work independently of work done in 1991 by Kelly and Davis [Kelly and Davis 1991]. Their core idea was exactly the same, namely using a GA to generate real valued weights to improve Knn performance, though the emphasis of the two groups is somewhat different (our work incorporates parallel processing and hidden feature processing, which is discussed in subsequent sections) ....

Kelly, James and Lawrence Davis, "Hybridizing the Genetic Algorithm and the K Nearest Neighbors Classification Algorithm", Proceedings of the 4th International Conference on Genetic Algorithms and their Applications, Morgan Kaufman Publishers, 1991.

....metric through the use of weighted attribute value differences. In case concept boundaries are not aligned with the attribute axes, rotations are necessary to fully profit from a weighted Euclidean distance measure. But determining rotation angles and weights also amounts to a search problem (cp. KD91] Note that our approach is independent of the orientation of concept boundaries w.r.t. attribute axes. To put it another way, the search problem remains the same whether concept regions are aligned with attribute axes or not. The instances of an optimal concept description are rotated the same ....

....a concept description becomes a striking advantage. The use of few, typical instances essentially amounts to an implicit generalization from examples that makes it easier to tolerate a remarkable level of noise as well as many near boundary instances or irrelevant attributes in the training set. KD91] propose an approach that does not construct new instances but transforms the given instances by rotations and attribute scalings which are optimized by a GA in order to support the k NN classifier. This idea is somehow complementary to our approach. But it is not evaluated using real world ....

J.D. Kelly and L. Davis. Hybridizing the genetic algorithm and the k nearest neighbors classification algorithm. In Proc. 4 th ICGA, San Diego, CA, USA, 1991.

.... to support neural network research in three different ways: the first is to select the input data or to transform the feature space, the second is to select a network learning rule and its parameters, and the third is to analyze a neural network [90, 119] Data preprocessing: Kelly and Davis [120] used GA to find rotations of data vectors and the scaling factors for each attribute, improving the performance of neural classifier. Other approaches are focused on data reduction. Frequently the data reduction improves the network performance and reduces the computing time. Chang and NEURAL ....

J.D. Kelly and L. Davis, Hybridizing the genetic algorithms and the k-nearest neighbors classification algorithms, In Fourth International Conference on Genetic Algorithms, Morgan Kaufmann, pp. 377-383, 1991.

.... selection and extraction, in combination with the k nearestneighbors classification rule, have been shown to provide increased accuracy over the knn rule alone, and can aid in the analysis of large datasets by isolating combinations of features that distinguish well among different pattern classes [4,5]. Genetic algorithms (GA s) have been applied to the problem of feature selection by Siedlecki and Sklanski [6] In their work, the genetic algorithm performs feature selection in combination with a knn classifier, which is used to evaluate the classification performance of each subset of features ....

....and a 0 indicating that it is omitted. The GA searches for a selection vector with a minimal number of 1 s, such that the error rate of the knn classifier remains below a given threshold. Later work by Punch et al. and Kelly Davis expanded this approach to use the GA for feature extraction [4,5]. 2 Instead of a selection vector consisting of only 0 s and 1 s, the GA manipulates a weight vector, in which a discretized real valued weight is associated with each feature. Prior to knn classification, the value of each feature is multiplied by the associated weight, resulting in a new set ....

J. D. Kelly and L. Davis, Hybridizing the Genetic Algorithm and the K Nearest Neighbors Classification Algorithm, Proc. Fourth Inter. Conf. Genetic Algorithms and their Applications (ICGA), pp. 377383, 1991.

....influences the computation of distance and consequently affects P ( x) Less important or completely irrelevant risk factors can have severely negative effects on prediction accuracy. The most obvious way to tackle the problem is to modify the source of it, namely the distance measure ffi (e.g. [15, 22]) For this purpose we generalize ffi as follows: ffi( x; z) v u u t n X i=1 c i Delta (x i Gamma z i ) 2 ; c i 0 Instead of the default values c i = 1, the coefficients c i can be chosen so as to reflect the importance of the respective risk factor i. The bigger c i is, the more ....

Kelly, J.D.; Davis, L.: Hybridizing the Genetic Algorithm and the K Nearest Neighbors Classification Algorithm, Proc. 4 th International Conference on Genetic Algorithms (ICGA-91), 1991, Morgan Kaufmann, pp. 377--383.

No context found.

J.D. Kelly and L. Davis. Hybridizing the genetic algorithm and the k nearest neighbors classification algorithm. In Proc. of the 4th International Conference on Genetic Algorithms and their Applications (ICGA'91), pages 377--383, 1991. 60

No context found.

J. D. Kelly and L. Davis, "Hybridizing the genetic algorithm and the k nearest neighbors classification algorithm," in Proceedings of the Fourth International Conference on Genetic Algorithms and their Applications, 1991, pp. 377--383.

No context found.

Kelly, J D and Davis, L `Hybridizing the Genetic Algorithm and the K nearest neighbors classification algorithm'. Proceedings of the Fourth International Conference on Genetic Algorithms, (1991) Morgan Kaufmann, San Mateo, Ca.

No context found.

James D. Kelly and Lawerence Davis. Hybridizing the Genetic Algorithms and the K Nearest Neighbors Classification Algorithm. In In Proceedings of Fourth International Conference on Genetic Algorithms, pages 377-383, 1991.

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