W. Siedlecki, J. Sklansky. A note on genetic algorithms for large-scale feature-selection. Pattern Recognition Letters, 1989.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....of genetic algorithm. It is called implicit parallelism and has been proved that it can set a lower bound of an N speedup over systematic sequential search [13] where N is the population size. In the last decade, genetic algorithms have been largely applied to the feature selection problem [37, 18, 30, 34]. The approach often combines di erent optimization objectives into a single objective function. The main drawback of this kind of strategy lies in the di culty of exploring di erent possibilities of trade o between classi cation accuracy and di erent subsets of selected features. In order to ....

....is based on the fact that several studies in the literature have been demonstrated that genetic algorithms would be more e ective than other methods when dealing with large scale feature selection (i.e. more than 50 features) For those readers interested in comparative studies, please see Refs. [7, 17, 34]. In order to show the robustness of the proposed strategy, we carry out experiments in two di erent contexts: isolated digits and string of digits. The latter is more di cult since we need to deal with problems such as fragmentation, overlapping and e ects of segmentation. Finally, we analyze ....

[Article contains additional citation context not shown here]

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large scale on feature selection. Pattern Recognition Letters, 10:335347, 1989.

....vector (Bitmap Encoding) a = a 1 ; a d ) where a i = 1 indicates the presence of the i th feature in the subset, while the absence of the i th feature is expressed by a i = 0. The bitmap encoding is well suited for an evolutionary feature selection technique based on a wrapper approach [12]. In [13] the bitmap encoding was generalized to a weighted encoding, where features were assigned different weights resulting in a modi cation (warping) of the feature space for a k NN classi er. In [14] bitmap encoding has been employed for the evolutionary search of feature subsets for an ....

Wojciech Siedlecki and Jack Sklansky. A Note on Genetic Algorithms for Large{Scale Feature Selection. Pattern Recognition Letters, 10:335{ 347, 1989.

....performance. GAs belong to the class of randomized heuristic search techniques, o#ering an attractive approach to feature subset selection. Although they have been used in various pattern recognition applications, their use in the area of computer vision is rather limited. Siedlecki et al. [15] has presented one of the earliest studies of GA based feature selection in the context of a K nearest neighbor classifiers. Roth et al. 16] have proposed extracting geometric features using GAs. Yang et al. 17] have also proposed using GAs for feature selection. Using several benchmark ....

W. Siedlecki and J. Sklansky, "A note on genetic algorithm for large-scale feature selection," Pattern Recognition Letter, vol. 10, pp. 335--347, 1989.

....the complete set of measurements Y = fy j jj = 1; 2; Dg to form feature set X k . The (k 1) st feature is then chosen from the set of available measurements, Y n X k , so that J(X k 1 ) max 8y j J(X k [ fy j g) y j 2 Y n X k : 3) 3.2. Genetic Algorithm based Selection (GAS) [12] In the basic approach to feature selection using EAs a feature subset is encoded as a binary vector (Bitmap Encoding) a = a 1 ; a d ) where a i = 1 indicates the presence of the i th feature in the subset, while the absence of the i th feature is expressed by a i = 0. To restrict the ....

Wojciech Siedlecki and Jack Sklansky. A Note on Genetic Algorithms for Large--Scale Feature Selection. Pattern Recognition Letters, 10:335--347, 1989.

....the set of transformed patterns. Using this information, the GA searches for a transformation that minimizes the dimensionality of the transformed patterns, while maximizing classification accuracy. A direct approach to using GAs for feature selection was introduced by Siedlecki and Sklanski [24]. In their work, a GA is used to find an optimal binary vector, where each bit is associated with a feature (Figure 3) If the i th bit of this vector equals 1, then the i th feature is allowed to participate in classification; if the bit is a 0, then the corresponding feature does not ....

W. Siedlecki and J. Sklansky, "A note on genetic algorithms for large-scale feature selection," Pattern Recognition Letters, vol. 10, pp. 335-347, 1989.

.... Sequential Search Methods [7] in which the number of forward and backtracking steps is dynamically controlled instead of being fixed before hand as in the plus l take away r algorithm (or (l; r) search [10] Other important contributions include the approaches based on Genetic Algorithms (GA) [9,3] which appear to offer an attractive alternative to heuristic tree search methods. However, none of these techniques has been thoroughly tested on large scale feature selection problems, involving hundreds rather than tens of features. The purpose of this paper is to investigate the applicability ....

....definition of the fitness function. The coding used to represent feature subsets consists of strings of D bits, ff 1 ; ff D , where ff i = 1 if the feature i is in the subset and ff i = 0 otherwise. This coding allows the use of all the standard genetic operators. The first GA approach [9] incorporates an appropriate penalty function to force the algorithm to search those feature subsets near to the feasibility boundary (threshold on the criterion function) Even though it was shown that this approach compares favourably with the (2,1) search for a 30 dimensional problem using ....

[Article contains additional citation context not shown here]

W. Siedlecki and J. Sklansky. A Note on Genetic Algorithm for Large-scale Feature Selection. Pattern Recognition Letters, 10(5):335--347, November 1989.

....identi ed by a particular method of coding the solutions into strings of some alphabet (usually binary) a particular form of the genetic operators adopted, and a particular de nition of the tness function. GAs have already been applied to a di erent formulation of the feature selection problem [8, 7]. As in this previous attempt, the problem considered herein is not in a form amenable to a direct GA optimization. Even though the space of solutions has a direct and natural translation into strings of bits, the de nition of an appropriate tness function and a meaningful crossover operator is ....

....quite similar but this method was faster than the other one for subsets with cardinality close to 30. 4 Conclusions and Further Work The use of GAs for feature subset search seems to be quite promising for dealing with high dimensional problems. The results reported here con rm the claims in [8] concerning the applicability of GAs to feature subset search problems and demonstrate their utility in the context of the Feature Selection problem formulation considered here which is distinct from the previous work reported in [8, 7] Despite this, there are still some open questions. First, ....

[Article contains additional citation context not shown here]

W. Siedlecki and J. Sklansky. A Note on Genetic Algorithm for Large-scale Feature Selection. Pattern Recognition Letters, 10(5):335-347, November 1989. 7

....run time inefficiency and system complexity. Since our two main sources of errors are related to classification problems (general purpose recognizer and lowlevel verifiers) we think that a plausible strategy to optimize the overall performance of the system lies in feature subset selection [10, 24]. A very popular approach to carry out this task in large scale problems is genetic algorithms [16] since the encoding of a feature subset into a chromosome is straightforward and the function that is optimized does not need to be smooth and can, therefore, be directly the classification accuracy ....

W.Siedlecki and J.Sklansky. A note on genetic algorithms for large scale on feature selection. Pattern Recognition Letters, 10:335--347, 1989.

....to data acquisition, its costs and difficulties associated with it. It seems natural, therefore, to seek a range of solutions rather than a single one. Such searches are more likely to be effectively performed by methods, which maintain a population of solutions, such as genetic algorithms (GA) [7] [8] GA input selection is often pursued by introducing different optimization Multiple Criteria Genetic Algorithms for Feature Selection in Neurofuzzy Modeling Christos Emmanouilidis 1 , Student Member, IEEE, Andrew Hunter, John MacIntyre, and Chris Cox, 1 ....

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large-scale feature selection. Pattern Recognition Letters. 10. 335-347, 1989.

....classification. Genetic algorithms offer a particularly attractive approach for this kind of problems since they are generally quite effective for rapid global search of large, non linear and poorly understood spaces. Moreover, genetic algorithms are very effective in solving large scale problems [16, 22]. This paper focuses on the feature subset selection for handwritten digit recognition through a modified wrapper based multi criterion approach using genetic algorithms in conjunction with a multi layer perceptron neural network. Two different versions of the genetic algorithm were explored: ....

W.Siedlecki and J.Sklansky. A note on genetic algorithms for large scale on feature selection. Pattern Recognition Letters, 10:335--347, 1989.

.... If the performance (on a test set) of an MLP is incidentally the best because of statistical fluctuations, one ends up exploring an inferior subset of features (Foroutan et al. 1987; Siedlecki et al. 1988) As a remedy, Siedlecky and Sklansky developed a genetic algorithm for feature selection (Siedlecki et al. 1989) and compared it with forward and backward search. Although the genetic algorithm outperformed both the forward, backward and branch and bound search schemes, their genetic approach is computationally much more complex. Their experiments showed that a backward search procedure yields close to ....

Siedlecki W., & Sklansky J. (1989). A note on genetic algorithms for largescale feature selection. Pattern Recognition Letters, 10 (5), 335--347.

.... information measures, error bounds, or empirical probability of error itself [9] can be such measures) Also a variety of algorithms have been developed over the past three decades that describe efficient methods to search the variables that contain discriminating information [10] 11] 12] [13], 3 [14] 26] Among parametric families, special attention in the literature on statistical pattern recognition has been paid to multivariate Gaussian densities. The most analytical results were obtained for a binary classification problem with data modeled to be multivariate Gaussian with ....

W. Siedlecki and J. Sklansky, "A Note on Genetic Algorithm for Large-Scale Feature Selection," Pattern Recognition Lett., vol. 10, 1989, pp. 335-347.

....8 others are based on minimization of the empirical error rate. 10 12 Most algorithms are suboptimal and do not guarantee the selection of the very best features. The branch and bound algorithm 11 is optimal only for a class of monotonic objective functions. Algorithms like those in [9] and [12] exhibit performance close to optimal. For performance comparison of di#erent algorithms see the paper by Jain and Zongker. 10 Many algorithms are based on the selection of singular features, 8 others are based on selection, consecutive inclusion and deletion of feature subsets (see ....

....and [12] exhibit performance close to optimal. For performance comparison of di#erent algorithms see the paper by Jain and Zongker. 10 Many algorithms are based on the selection of singular features, 8 others are based on selection, consecutive inclusion and deletion of feature subsets (see [9] and [12] for example) In this paper, we consider a recognition problem with M, M # 2, classes (or populations) We assume that the populations belong to the same parametric class but di#er in the values of their parameters. For the analysis we assume that the entries in the measurement ....

W. Siedlecki and J. Sklansky, "A Note on Genetic Algorithm for Large-Scale Feature Selection," Pattern Recognition Lett., vol. 10, 1989, pp. 335-347.

....helps the wrapper feature selection process to select only relevant variables and to, e ectively and eciently, reduce the complexity of the classi cation problem. The feature selection problem is an optimisation problem with restrictions that has been solved by means of GAs in di erent proposals ([50, 8, 3, 54]) In the next section, we describe our genetic proposal for the wrapper feature selection process. 5.3 Steady State GAs for Feature Selection The rst stage of the learning process is carried out by means of a feature selection algorithm based on a GA with a variant of the pure steady state ....

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large-scale feature selection. Pattern Recognition Letters, 10:335-347, 1989.

....representing an object with its feature vector is a fundamental and crucial step. Although there are a number of methods for automatic feature generation (e.g. Kudenko and Hirsh, 1998] in most cases features are provided by the user. Machine support is available for feature selection (e.g. [Siedlecki and Sklansky, 1989; Yang and Honavar, 1997] and feature extraction (e.g. Sherrah et al. 1997] While the user is indispensable if objects are not accessible at all for the computer (e.g. tumour cells , persons ) an automated feature generation, other than feature extraction which depends on a given set of ....

Siedlecki, W. and Sklansky, J. (1989). A note on genetic algorithms for large-scale feature selection. Pattern Recognition Letters, 10:335--347.

....extraction algorithms based on the wrapper model are characterised by the construction of the Classification System by means of an inductive learning process, for the evaluation of the variable subsets. In this type of algorithms, we should highlight the works carried out by Siedlecki and Sklansky [71], Brill et al. 9] Punch et al. 68] Ray et al. 70] and Yang and Honavar [79] Siedlecki and Sklansky [71] introduce the use of GAs for the selection of features in the design of Classification Systems. For this aim, they propose a binary coded GA, in which a 0 value represents the absence ....

....by means of an inductive learning process, for the evaluation of the variable subsets. In this type of algorithms, we should highlight the works carried out by Siedlecki and Sklansky [71] Brill et al. 9] Punch et al. 68] Ray et al. 70] and Yang and Honavar [79] Siedlecki and Sklansky [71] introduce the use of GAs for the selection of features in the design of Classification Systems. For this aim, they propose a binary coded GA, in which a 0 value represents the absence of the feature and an 1 value its presence. The fitness function combines the error measure obtained from the ....

[Article contains additional citation context not shown here]

Siedlecki, W. and Sklansky, J. (1989), "A note on genetic algorithms for large-scale feature selection," Pattern Recognition Letters, Vol. 10, pp. 335-347.

....size, crossover and mutation probabilities, and number of generations also play an important role in obtaining good quality results using GAs. Genetic algorithms have been used in many pattern recognition and image processing applications including image segmentation [1] feature selection [11], and shape analysis [2] The main drawback of genetic algorithms is the amount of time taken for convergence. The search space grows exponentially as a function of the problem size. Hence, the number of generations needed to reach a global solution increases rapidly. A number of methods have been ....

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large-scale feature selection. Pattern Recognition Letters, 10(11):335--346, November

....vector (Bitmap Encoding) a = a 1 ; a d ) where a i = 1 indicates the presence of the i th feature in the subset, while the absence of the i th feature is expressed by a i = 0. The bitmap encoding is well suited for an evolutionary feature selection technique based on a wrapper approach (Siedlecki and Sklansky, 1989). In (Punch et al. 1993) the bitmap encoding was generalized to a weighted encoding, where features were assigned di erent weights resulting in a modi cation (warping) of the feature space for a k NN classi er. In (Yang and Honavar, 1997) bitmap encoding has been employed for the evolutionary ....

Siedlecki, W. and Sklansky, J. (1989). A Note on Genetic Algorithms for Large{Scale Feature Selection. Pattern Recognition Letters, 10:335-347.

....alphabet (usually a binary alphabet) The searching of this representation space is performed using so called Genetic Algorithms (GA s) The genetic algorithm is now widely recognized as an effective search paradigm in many areas. It has been used in engineering applications such as: clustering [7,16,18] and pipeline optimization [9] In the context of design, GA s have been used for: VLSI cell placement [12,13] floor plan design [14] air injected hydrocyclone [11] cache design [1] network design [6] and others. The classes of problems encountered in design include many that are not easily ....

W. Siedlecki and J. Sklansky, "A Note on Genetic Algorithms for Large-scale Feature Selection, " Pattern Recognition Letters, October 1989, pp. 335-347.

.... and this led to the development of floating search methods [4] Approaches that maintain a population of solutions, such as evolutionary algorithms (EAs) are less likely to be restricted by interdependencies among features and may speedily perform efficient searches in high dimensional spaces [5]. Evolutionary algorithms have been used many times to aid the selection of feature subsets in various classification tasks (e.g. 6] 7] This work introduces a multiobjective evolutionary algorithm (MOEA) setting for the feature selection problem. Multiobjective genetic algorithms have ....

W. Siedlecki and J. Sklansky, (1989) "A note on genetic algorithms for large-scale feature selection," Pattern Recognition Letters, vol. 10, pp. 335-347.

....[649] Optics Letters, 300, 401, 621] Optimization (UK) 917] ORSA Journal on Computing, 40, 128] Pattern Recognit. Image Anal. Russia) 249, 140] Pattern Recognition, 143, 144, 535, 541, 278, 576, 344, 382, 396, 592, 413, 596, 111, 435, 599, 603] Pattern Recognition Letters, [167, 50, 57, 761, 58, 221, 67, 784, 543, 241, 244, 69, 252, 364, 402, 876, 406, 595, 601, 452, 496, 505] Power Syst. Technol. China) 686] Proceedings of IEE Vision, Image Signal Processing, 139] R. F. Des. USA) 858] Real Time Systems , 849] Res. Rep. Kogakuin Univ. Japan) 245] Seimitsu Kogaku Kaishi, 1111] Sens. Actuators A. Phys. Switzerland) 902] Signal Processing, 390, ....

....[702] Shimojima, K. 1117] Shin, Hyundoo, 966] Shine, J. A. 55] Shinohara, K. 245] Shioyama, T. 243] Shiozawa, M. 768] Shirakawa, H. 321] Shiraki, H. 296] Shmerko, V. 536] Shyu, Ming Suen, 213, 435] Si, A. 276] Sidla, Oliver, 1058] Sidla, O. 1054] Siedlecki, W. [505, 628] Sikora, Riyaz, 40] Sim, H. C. 569, 578] Simon, Dan, 732] Simpson, Angus R. 940] Simpson, Marc T. 21, 22] Sims, S. Richard F. 501] Singh, Montek, 596] Singh, R. 823] Singher, Liviu, 687] Sirin, Izzet, 136] Sizmann, R. 709] Sklansky, Jack, 576] Sklansky, J. 546, 505, 628] ....

[Article contains additional citation context not shown here]

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large scale feature selection. Pattern Recognition Letters, 10(5):335--347, November 1989. ga:Sklansky89a.

....[311] Optics Communications, 477] Optics Letters, 484, 489, 498, 325, 530] Pattern Recognit. Image Anal. Russia) 153, 14] Pattern Recognition, 20, 540, 541, 23, 24, 109, 557, 186, 584, 258, 299, 318, 593, 340, 595, 368, 600, 388, 389, 390, 405] Pattern Recognition Letters, [51, 86, 116, 124, 559, 146, 149, 158, 280, 326, 333, 594, 606, 398, 448, 458] Pattern Recognition. Lett. Netherlands) 598] Proc. Inst. Mech. Eng. I, J. Syst. Control Eng. UK) 332] Proceedings of IEE Vision, Image Signal Processing, 12] Res. Rep. Kogakuin Univ. Japan) 150] Scienti c Computing World, 473] Seimitsu Kogaku Kaishi, 708] Signal Process Image ....

....Shine, J. A. 97] Shinohara, K. 150] Shioyama, T. 148] Shirakawa, H. 234] Shiraki, H. 205] Shmerko, V. 120] Shoenauer, Marc, 482] Shyu, Ming Suen, 113, 368] Si, A. 184] Sidla, Oliver, 671] Sidla, O. 668] Sidorowich, John J. 497] Sieber, I. 514] Siedlecki, W. [458, 627] Sim, H. C. 579, 279] Sims, S. Richard F. 454] Singh, Montek, 595] Singher, Liviu, 508] Sizmann, R. 535, 536] Skaar, J. 515] Sklansky, Jack, 584] Sklansky, J. 562, 458, 627] Skolnick, Michael M. 459] Slate, D. J. 611] Sluzek, A. 169] Smith, Jim, 649] Smith, M. I. ....

[Article contains additional citation context not shown here]

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large scale feature selection. Pattern Recognition Letters, 10(5):335-347, November 1989. ga:Sklansky89a.

.... features which do poorly alone but offer valuable information together [1] Approaches that maintain a population of solutions, such as genetic algorithms (GAs) are more likely to speedily perform efficient searches in high dimensional spaces, with strong interdependencies among the features [18]. A feature subset is represented as a bit string, with the setting of each bit indicating whether the corresponding feature is used, or not. Yet, even for single objective problems, GAs can prematurely converge to sub optimal solutions. This can be due to the existence of super fit individuals, ....

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large-scale feature selection. Pattern Recognition Letters. 10. 335-347, 1989.

.... cation accuracy (direct error minimization) Classi cation accuracy error is a notoriously dicult optimization criterion, making feature selection the most challenging task in pattern recognition [45] 46] 47] Although many authors advocate feature selection by GAs [48] 49] 50] 51] [52], others are skeptical [53] and warn that the results are often not as good as expected, compared with other (simpler ) feature selection algorithms [54] Kuncheva [22] uses GAs to select features for the individual classi ers in a classi er fusion scheme. The subset of features used by an ....

....sequential backward selection (SBS) The error of the classi er (chosen for that data set ) was used as the criterion function J . The procedure starts with the whole set of features (of cardinality n) and discards one feature at each step as shown in Figure 4. 2. A GA for feature selection [54] [52]. The GA described in Section 3 is used, where the tness function is the classi cation accuracy (1 the error rate) using the feature subset represented as a chromosome. The chromosome is a binary string of length n. The ith bit takes values 0 if the feature is not in the subset or 1 if the ....

W. Siedlecki and J. Sklansky, \A note on genetic algorithms for large-scale feature selection," Pattern Recognition Letters, vol. 10, pp. 335-347, 1989.

....optimality. For instance, algorithms such as sequential search [30, 19] branch and bound [26] nonlinear optimization [5] and simulated annealing [27] have been applied. The formulation of feature selection as a combinatorial optimization problem has also lead to the use of genetic algorithms [28, 31]. A recent review of these methods can be found in [8] Regardless of the search algorithm employed, most previous methods evaluated potential solutions in terms of predictive accuracy. Speci cally, the data set could be divided into training and test sets, with the error rate on the test set used ....

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large-scale feature selection. Pattern Recognition Letters, 10:335-347, 1989.

....time a GA can find near optimal maxima of a fitness function, even for very poorly behaved functions in high dimensional spaces. As such, its applicability to the optimum feature subset selection problem is apparent, and a number of authors have investigated the use of GAs for feature selection [15, 16, 10, 17]. GAs have been used in combination with neural nets in several ways. They have been used to select feature sets for classification by a neural net trained using conventional learning algorithms (eg [3, 6] and to evolve the weight vector and or architecture of a neural net (for a review see ....

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large-scale feature selection. Pattern Recognition Letters, 10(5):335--347, 1989.

....selection one may proceed as follows. For any subset of variables, a classification rule is constructed using densities p i (x) 1 i g) and its performance (misclassification error) is evaluated modelfree by leave one out. This can be either done by exhaustive search or by genetic algorithms (Siedlecki and Sklanski (1989)) which implement a more intelligent strategy inspired by evolution. With more than 20, say, variables, exhaustive search is not feasible. With variables selection the important variables are detected directly and no further interpretation is needed. The simplest group densities p i (x) are ....

W. Siedlecki und J. Sklansky (1989), A note on genetic algorithms for large-- scale feature selection, Pattern Recognition Letters, 10, 335-347 .

....evolution. A GA can find near global optimal parameter values even for poorly behaved functions, given sufficient time. As such, its applicability to the optimum feature subset selection problem is apparent, and a number of authors have investigated the use of GAs for feature selection [8, 10, 4, 11]. The actual subset of features which is optimal for a given problem will depend upon the classifier used. In many real world problems, classes are not linearly separable, and a non linear classifier, such as a neural net, is required. GAs have been used in combination with neural nets in several ....

Siedlecki, W. & Sklansky, J.(1989). A note on genetic algorithms for large-scale feature selection. Pattern Recognition Letters 10(5): 335--347.

.... are said unfaithful if they have the same feature values but different class labels) Several authors have explored the use of randomized population based heuristic search techniques such as genetic algorithms (GA) for feature subset selection for decision tree and nearest neighbor classifiers [Siedlecki and Sklansky, 1989; Brill et al. 1992; Punch et al. 1993; Richeldi and Lanzi, 1996] or rule induction systems [Vafaie and De Jong, 1993] A related approach used lateral feedback networks [Guo, 1992; Kothari and Agyepong, 1996] to evaluate feature subsets [Guo and Uhrig, 1992] Feature subset selection techniques ....

Siedlecki, W. and Sklansky, J. (1989). A note on genetic algorithms for large-scale feature selection.

....for an accuracy of 41:18 . We found we can improve the classification accuracy through feature selection techniques. Some effective conventional methods for feature selection are sequential forward selection [5, 11] sequential floating feature selection [12] and genetic algorithm search [13, 14]. Sequential forward selection achieved 17 85 errors, or 80 accuracy by selecting 13 features. These features were engin, action, david, contempl, affirm, architectur, ave, osha, abund, rehabilit, notic, commerc, transact. The genetic algorithm feature selection method used a fixed size subset ....

W. Siedlecki and J. Sklansky, "A note on genetic algorithms for large scale feature selection," Pattern Recognition Letters, vol. 10, pp. 335--347, November 1989.

....done with different search algorithms. Branch and bound, besides exhaustive search, is the only known search algorithm for feature selection that guarantees to give an optimal solution. This algorithm was introduced by Fukunaga and Narendra in [10] According to W. Siedlecki and J. Sklansky, see [22] for details) optimal search strategies are likely to become non polynomial. They suggest to use branch and bound if the dimension of the original feature space is below 20. In our case we deal with large scale feature selection. The initial number of features is beyond two hundred and we need a ....

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large scale feature selection. Pattern Recognition Letters, 10:335--347, 1989.

.... 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 selected by the GA. The GA maintains a feature selection vector consisting of a single bit for each feature, ....

W. Siedlecki and J. Sklansky, A Note on Genetic Algorithms for Large-Scale Feature Selection, Pattern Recognition Letters, vol. 10, pp. 335-347, 1989.

.... 1992; Modrzejewski, 1993; Liu Setiono, 1995; John et al. 1994; Kohavi, 1994; Kohavi Frasca, 1994; Koller Sahami, 1996] Others have explored randomized [Liu Setiono, 1996b; Liu Setiono, 1996a] and randomized population based heuristic search techniques such as genetic algorithms (GA) [Siedlecki Sklansky, 1989; Punch et al. 1993; Vafaie De Jong, 1993; Brill et al. 1992; Richeldi Lanzi, 1996] to select feature subsets for use with decision tree or nearest neighbor classifiers. Feature subset selection algorithms can be classified into two categories based on whether or not feature selection is ....

Siedlecki, W., & Sklansky, J. (1989). A Note on Genetic Algorithms for Large-scale Feature Selection. IEEE Transactions on Computers, 10, 335--347.

....and Sklansky [6] RBAB aims to find the smallest subset for which the criterion value is not under a given threshold and the search is carried out for a larger set of subsets for which the criterion values are over Gamma ffi(ffi 0) where ffi is called a margin. GA: The genetic algorithm [7, 8]. In GA, a feature subset is represented by a binary string with length n, called a chromosome, with a zero or one in position i denoting the absence of presence of feature i. Each chromosome is evaluated in its fitness through an optimization function in order to survive to the next generation. ....

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large-scale feature selection. Pattern Recognition Letters, 10:335--347, 1989.

No context found.

W. Siedlecki, J. Sklansky. A note on genetic algorithms for large-scale feature-selection. Pattern Recognition Letters, 1989.

No context found.

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large scale on feature selection. Pattern Recognition Letters, 10:335--347, 1989.

No context found.

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large scale on feature selection. Pattern Recognition Letters, 10:335--347, 1989.

No context found.

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large scale on feature selection. Pattern Recognition Letters, 10:335347, 1989.

No context found.

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large-scale feature selection. Pattern Recognition Letters, 10:335--347, 1989. 58, 60, 62, 84, 85

No context found.

W. Siedlecki and J. Sklansky, "A note on genetic algorithms for large scale on feature selection," Patt. Recogn. Lett. 10 (1989) 335--347.

No context found.

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large scale on feature selection. Pattern Recognition Letters, 10:335--347, 1989.

No context found.

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large scale on feature selection. Pattern Recognition Letters, 10:335347, 1989.

No context found.

W. Siedlecki and J. Sklansky, "A note on genetic algorithms for large-scale feature selection," Pattern Recognit. Lett., vol. 10, pp. 335-347, 1989.

No context found.

W. Siedlecki and J. Sklansky, "A note on genetic algorithm for large-scale feature selection," Pattern recognition letter, vol. 10, pp. 335-- 347, 1989.

No context found.

W. Siedlecki and J. Sklansky. A note on genetic algorithm for large-scale feature selection. Pattern Recognition Letters, 10(11):335--347, 1989.

No context found.

W. Siedlecki and J. Sklansky. A note on genetic algorithms for large scale on feature selection. Pattern Recognition Letters, 10:335347, 1989.

No context found.

W. Siedlecki and J. Sklansky, "A note on genetic algorithms for large-scale feature selection," Pattern Recognition Letters, vol. 10, pp. 335--347, 1989.

No context found.

W. Siedlecki and J. Sklansky, "A note on genetic algorithms for large scale feature selection", Pattern Recogn Lett, 10, 335-47, (1989).

No context found.

Siedlecki, W. & Sklansky, J. 1989. 'A note on genetic algorithms for large-scale feature selection', Pattern Recognition Letters vol. 10, no. 5, p.335 - 347.

No context found.

W. Siedlecki and J. Sklansky, \A note on genetic algorithms for large-scale feature selection, " Pattern Recognition Letters, vol. 10, pp. 335-347, 1989.

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