Research And Implementation Of The Integration Of Min-Max Modular SVM And Random Subspace

Min-Max Modular Support Vector Machine (M3-SVM) is a kind of effective ensemble methodto deal with large-scale data classification problem. In order to improve the performance ofM3-SVM for large-scale unbalanced data classification problem, the M3-SVM algorithm iscombined with random subspace to increase the diversity between base classifiers and improve theperformance of base classifier. At the same time, the parallel M3-SVM is developed based on MPI.The concrete works are introduced as follows:On the one hand, this thesis proposes a M3-SVM algorithm based on random subspace. Thismethod selects different feature subsets for base classifier to improve on the diversity between thebase classifier and improve the classification performance of Min-Max Modular network.On the other hand, the parallel M3-SVM algorithm is developed based on Message PassingInterface-MPI. M3network can make full use of the distributed computing systems to solvelarge-scale data classification problem. In our case, the M3-SVM algorithm is combined with theMPI parallel environment to enhance the efficiency of the algorithm.The experiments on real-world data sets including unbalanced data indicate that the proposedrandom subspace strategy can enhance the classification performance of M3-SVM and enlarge thediversity between base classifiers in M3-SVM. The experiments on MPI show that the algorithmparallelization can significantly improve the efficiency of the original algorithm.