TY - JOUR T1 - Support Vector Random Machines (SVRMs), A OptimumMulticlassifier for Big Data TT - ماشین‌های تصادفی بردار پشتیبان، طبقه‌بندی دسته‌جمعی بهینه داده‌های با ابعاد بالا JF - kntu-jgit JO - kntu-jgit VL - 5 IS - 1 UR - http://jgit.kntu.ac.ir/article-1-428-en.html Y1 - 2017 SP - 133 EP - 152 KW - Support Vector Machine (SVM) KW - Ensemble method KW - Feature space KW - Bootstarp KW - Aggregation N2 - Although, the distinction between the land cover classes was increased in large feature space of remote sensing images, but the low number of training data prevent this. In order to solve this problem, ensemble classification methods can be used instead of individual classifiers. In this paper, a new method for ensemble support vector machine was proposed called “Support Vector Random Machines (SVRMs(”. In proposed method, bootstrap was produced using modification of training data and feature space. Simultaneous boosting SVM was used for basic classifiers. Then, classification map was resulted using SVM fusion of basic classifier. Hyperspectral and Polarimetric SAR data was chosen for evaluation performance of the SVRMs. Experiments were evaluated from three different points of view: First, evaluation against other ensemble SVM methods; second, evaluation against various feature selection methods in classification and third, evaluation against the various basic and new classification methods. As the results, proposed method is 16% better than the individual SVM classifier in hyperspectral data and this is 10% in PolSAR data. Also, the classification results of SVRMs in various classes compared to other SVM ensemble method were improved. The results reported from the proposed method compared to the other feature selection method (Genetic Algorithm) has the effectual performance in classification. The results show that the proposed method presents a better performance compared to the basic classification methods (maximum likelihood and wishart) and advanced classification (random forest and neural network). M3 10.29252/jgit.5.1.133 ER -