Original scientific paper
https://doi.org/10.20532/cit.2016.1003342
Microcanonical Annealing and Threshold Accepting for Parameter Determination and Feature Selection of Support Vector Machines
Seyyid Ahmed Medjahed
; Université des Sciences et de la Technologie d'Oran Mohamed Boudiaf, Algérie
Tamazouzt Ait Saadi
; University of Le Havre, France
Abdelkader Benyettou
; Université des Sciences et de la Technologie d'Oran Mohamed Boudiaf, Algérie
Mohammed Ouali
; Computer Science Department, University of Sherbrooke, Canada
Abstract
Support vector machine (SVM) is a popular classification technique with many diverse applications. Parameter determination and feature selection significantly influences the classification accuracy rate and the SVM model quality. This paper proposes two novel approaches based on: Microcanonical Annealing (MA-SVM) and Threshold Accepting (TA-SVM) to determine the optimal value parameter and the relevant features subset, without reducing SVM classification accuracy. In order to evaluate the performance of MA-SVM and TA-SVM, several public datasets are employed to compute the classification accuracy rate. The proposed approaches were tested in the context of medical diagnosis. Also, we tested the approaches on DNA microarray datasets used for cancer diagnosis. The results obtained by the MA-SVM and TA-SVM algorithms are shown to be superior and have given a good performance in the DNA microarray dana sets which are characterized by the large number of features. Therefore, the MA-SVM and TA-SVM approaches are well suited for parameter determination and feature selection in SVM.
Keywords
support vector Machines; microcanonical annealing; threshold acceptance; parameter determination; feature selection
Hrčak ID:
173013
URI
Publication date:
22.12.2016.
Visits: 900 *