J. M. Pena, J. A. Lozano, and P. Larranaga. An empirical comparison of four initialization methods for the k-means algorithm. Pattern Recognition Lett., 20:1027--1040, 1999.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....in the data set. A popular clustering method that minimizes the clustering error is the k means algorithm. However, the k means algorithm is a local search procedure and it is wellknown that it su ers from the serious drawback that its performance heavily depends on the initial starting conditions [2]. To treat this problem several other techniques have been developed that are based on stochastic global optimization methods (e.g. simulated annealing, genetic algorithms) However, it must be noted that these techniques have not gained wide acceptance and in many practical applications the ....

J. A. L. J. M. Pena and P. Larranaga, \An empirical comparison of four initialization methods for the k-means algorithm," Pattern Recognition Letters, vol. 20, pp. 1027-1040, 1999.

....the k meansalgnsbxD is a local search procedure and it is well known that it su#ers from the serious drawback that its performance heavily depends Corresponding author. Tel. 30 6510 98810; fax: 306510 98889. E mail address: arly cs.uoi.g (A. Likas) on the initialstarting conditions [2]. To treat this problem several other techniques have been developed that are based on stochasticgtoch optimization methods(e.g simulated annealing gnneal algalingDE However, it must be noted that these techniques have notgtb#D wide acceptance and in many practical applications ....

J.A. Lozano, J.M. Pena, P.Larranag# An empirical comparison of four initialization methods for the k-means algsbqDEq PatternRecogb#SSC Lett. 20 (1999) 1027--1040.

....as well as instance order. We have therefore chosen to use the Partition Around Medioids initialization as proposed by Kaufman [7] This initialization was (empirically) found to be the best of four classical initialization methods when looking at e ectiveness, robustness and convergence speed [13]. For each clustered continuous predictor variable we have now k clusters and their mediods. We can now fuzzify the clusters by de ning membership functions for each fuzzy cluster. We will use three di erent types of membership functions as de ned in De nition 1. De nition 1. Let F = fF 1 ; ....

J. Pena, J. Lozano, and P. Larranaga. An empirical comparison of four initialization methods for the k-means algorithm, 1999.

....So this new approximation can be used in problems, as exploratory data analysis, where data are represented by strings. Depending on the initialization of the k median algorithm the behaviour is di#erent. Some di#erent initializations of the algorithm have been proposed in the literature ( 1][8]) In this work we have used the most simple initialization that consists in selecting randomly k cluster representatives. In the next section this approximated median algorithm is described. Some experiments using synthetic and real data to compare the approximated and the exact set median are ....

Pena, J.M., Lozano, J.A., Larranaga, P.: An empirical comparison of four initialization methods for the K--means algorithm. Pattern Recognition Letters 20 1027--1040 (1999).

....almost always get trapped somewhere close to the initial starting configuration. In other words, it is di#cult to sample through a large configuration (parameter) space. The conventional approach is to do a large number of runs with random initial starts and pick up the best one as the result [24, 26]. Besides random starts, there are a number of initialization methods, most of which concentrate on how to intelligently choose the starting configurations (the K centers) in order to be as close to the global minima as possible [5, 25, 22, 17] However, these approaches are limited by the ....

J. Pena, J. Lozano, and P. Larranaga. An empirical comparison of four initialization methods for the k-means algorithm. Pattern Recognition Letters, 50:1027--1040, 1999.

....in the data set. A popular clustering method that minimizes the clustering error is the k means algorithm. However, the k means algorithm is a local search procedure and it is wellknown that it su#ers from the serious drawback that its performance heavily depends on the initial starting conditions [2]. To treat this problem several other techniques have been developed that are based on stochastic global optimization methods (e.g. simulated annealing, genetic algorithms) However, it must be noted that these techniques have not gained wide acceptance and in many practical applications the ....

J. A. L. J. M. Pena and P. Larranaga, "An empirical comparison of four initialization methods for the k-means algorithm," Pattern Recognition Letters, vol. 20, pp. 1027--1040, 1999.

....the data set. A popular clustering method that minimizes the clustering error is the k means algorithm. However, the k means algorithm is a local search procedure and it is well known that it su ers from the serious drawback that its performance heavily depends on the initial starting conditions [2]. To treat this problem several other techniques have been developed that are based on stochastic global optimization methods (e.g. simulated annealing, genetic algorithms) However, it must be noted that these techniques have not gained wide acceptance and in many practical applications the ....

J. A. Lozano J. M. Pena and P. Larranaga, \An empirical comparison of four initialization methods for the k-means algorithm," Pattern Recognition Letters, vol. 20, pp. 1027-1040, 1999.

No context found.

J. M. Pena, J. A. Lozano, and P. Larranaga. An empirical comparison of four initialization methods for the k-means algorithm. Pattern Recognition Lett., 20:1027--1040, 1999.

No context found.

J. M. Pena, J. A. Lozano and P. Larranaga. An empirical comparison of four initialization methods for the k-means algorithm. Pattern Recognition Letters, 20, 1027--1040, 1999.

No context found.

J.M. Pe~na, J.A. Lozano, and P. Larra~naga. An empirical comparison of four initialization methods for the k-means algorithm. Pattern recognition letters, 20:1027-1040, 1999.

No context found.

J. Pe~na, J. Lozano, and P. Larra~naga. An empirical comparison of four initialization methods for the k-means algorithm. Pattern recognition letters, 20:1027-1040, 1999.

No context found.

J. M. Pena, J. A. Lozano, and P. Larranaga, "An empirical comparison of four initialization methods for the k--means algorithm," Pattern Recognition Letters 20, pp. 1027--1040, July 1999.

No context found.

J. M. Pena, J. A. Lozano, and P. Larranaga, "An empirical comparison of four initialization methods for the k--means algorithm," Pattern Recognition Letters 20, pp. 1027--1040, July 1999.

No context found.

J. M. Pena, J. A. Lozano, and P. Larranaga, "An empirical comparison of four initialization methods for the k--means algorithm," Pattern Recognition Letters 20, pp. 1027--1040, July 1999.

No context found.

J.M. Pena, J.A. Lozano, and P. Larranaga. An empirical comparison of four initialization methods for the K-Means algorithm. Pattern Recognition Letters, 20:1027--1040, 1999.

No context found.

Pen~a, J.M., Lozano, J.A. and Larran~aga P.: "An empirical comparison of four initialization methods for the k-means algorithm", Pattern Recognition Letters, 20, 1999, 1027-1040. 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