Citation Number
1
Views
26
Downloands
2
Kümeleme uygulamalarında en sık kullanılan algoritmalardan biri olan k-ortalamalar yönteminin tatbik edilmesinde karşılaşılan başlıca zorluk, gözlem sayısına bağlı olarak hesaplama karmaşıklığının artması ve problem için küresel en iyi çözüme yakınsayamamadır. Üstelik problem boyutunun ve karmaşıklığının artması halinde k-ortalamalar yönteminin performansı daha da kötüleşmektedir. Tüm bu nedenlerden ötürü klasik k-ortalamalar prosedürü yerine daha hızlı ve başarılı bir kümeleme algoritması geliştirme çalışmaları önem kazanmaktadır. Meta-sezgisel kümeleme (MSK) algoritmaları bu amaçla geliştirilmişlerdir. MSK algoritmaları sahip oldukları arama yetenekleri sayesinde karmaşık kümeleme problemlerinde yerel çözüm tuzaklarından kurtulabilmekte ve küresel çözüme başarılı bir şekilde yakınsayabilmektedirler. Bu makale çalışmasında literatürde yer alan güncel ve güçlü meta-sezgisel arama (MSA) teknikleri kullanılarak MSK algoritmaları geliştirilmekte ve performansları karşılaştırılarak en etkili yöntem araştırılmaktadır. Bu amaçla güncel ve güçlü MSA teknikleri ile k-ortalamalar yöntemi melezlenerek 10 farklı MSK algoritması geliştirilmiştir. Geliştirilen algoritmaların performanslarını ölçmek için 5 farklı kümeleme veri seti kullanılmıştır. Deneysel çalışmalardan elde edilen veriler istatistiksel test yöntemleri kullanılarak analiz edilmiştir. Analiz sonuçları, makalede geliştirilen MSK algoritmaları arasında AGDE tabanlı yöntemin hem yakınsama hızı hem de küresel optimum çözüme yakınsama miktarı açısından kümeleme problemlerinde rakiplerine kıyasla üstün bir performansa sahip olduğunu göstermektedir.
The main challenge faced in the practice of the method of k-mediates, one of the algorithms most commonly used in aggregation applications, is the increased complexity of calculation depending on the number of observations and the failure to approach the best global solution to the problem. In addition, with the increase in the size and complexity of the problem, the performance of the k-mediates method is even worse. For all of these reasons, the work on developing a faster and more successful aggregation algorithm instead of the classic k-mediates procedure becomes important. Metaphysical aggregation (MSK) algorithms have been developed for this purpose. The MSK algorithms are able to get rid of local solutions traps in complex aggregation problems and successfully approach global solutions. In this article, the MSK algorithms are developed using the current and powerful meta-sensitive search (MSA) techniques included in literature and the most effective method of comparing performance is researched. For this purpose, 10 different MSK algorithms have been developed with the up-to-date and powerful MSA techniques and the method of k-mediates. 5 different aggregate data sets have been used to measure the performance of the developed algorithms. The data obtained from experimental studies were analyzed using statistical testing methods. The results of the analysis show that among the MSK algorithms developed in the article, the AGDE-based method has superior performance compared to its competitors in accumulation problems in terms of both the approximation speed and the global optimum solution approximation quantity.
One of the most frequently used algorithms in clustering analysis, the main difficulty encountered in applying the k-means method is that the calculation complexity increases due to the number of observations and it cannot converge to the global best solution for the problem. Moreover, if the problem size and complexity increases, the performance of the k-means method gets worse. For all these reasons, it is important to develop a faster and successful clustering algorithm instead of the classical k-means procedure. Meta-heuristic clustering (MSK) algorithms have been developed for this purpose. Thanks to their search capabilities, MSK algorithms can get rid of local solution traps in complex clustering problems and successfully converge to the global solution. Therefore, the cluster success of MSK methods is directly affected by the search success of MSA techniques. In this article, MSK methods are developed by using current and powerful MSA techniques in the literature and the most effective method is investigated by comparing the performance of these algorithms. For this purpose, ten different MSK algorithms have been developed by hybridizing the k-means method with current and powerful MSA techniques. Five different clustering data sets were used to measure the performance of the developed algorithms. Data obtained from experimental studies were analyzed using statistical test methods. The results of the analysis show that among the MSK algorithms developed in the article, the AGDE-based method has a superior performance compared to its competitors in cluster problems in terms of both the convergence rate and the amount of convergence to the global optimum solution.
Field : Mimarlık, Planlama ve Tasarım; Mühendislik
Journal Type : Uluslararası
Article Detail 
Citation Owners
Similar Articles
Download PDF
Journal Information
Listen Paper
Print This Page
Share
Mendeley
