An Efficient Learning Method Using a Distributed Support Vector Machine Based on Controlling Data Transger

Abstract

The developments in information technology have highlighted the importance of analyzing big data stored in various databases. With this as a background, the importance of distributed data mining (DDM), which is the technique of implementing data mining while databases are not transmitting raw data to each other, has been advocated. As one of the methods, Forrero et al. proposed the method of optimal learning with a support vector machine (SVM) that uses the alternating direction method of multipliers (ADMM) in the context of DDM. The apparatus is called a consensus-based distributed support vector machine (D-SVM). This method can learn the optimal hyperplane with a relatively small number of iterations and minimal communication cost for an arbitrary network structure without sharing data. However, when the statistical characteristics of data stored in each database are quite different, this method requires many iterations until convergence. Needless to say, it is better that the number of iterations and total communication cost for the learning classifier are minimized. In this study, we propose a new and effective learning method that reduces the number of iterations considering the network structure, provided that all of the nodes are connected to each other. To verify the effectiveness of the proposed method, a simulation experiment using the UCI machine learning repository and artificial data is conducted.