Leveraging Parallel Computing for Efficient Cascade and Distributed Support Vector Machines on Large-Scale Data

The Support Vector Machine (SVM) is a popular classification algorithm and is known for its generalization properties in machine learning. In this paper, we propose parallel implementation of Cascade and Distributed SVM algorithms for large scale dataset and demonstrate the performance enhancement of the algorithm for large-scale problems including training time, scalability, error ratio, accuracy, etc. The major contribution of the paper is distributed optimization of the SVM algorithm through the parallelism methods and openMPI and openMP techniques. First, we distributed data among nodes in the cluster with a same size of each data file. Second parallel and distributed computing of SVM for incremental dataset to show significant enhancement in the computation time. We apply parallel Cascade and Distributed SVM to various real-world big data classification problems and compare them with state-of-the-art SVM algorithms implemented in python-open MPI on public data sets namely IJCNN, Webspam. The observation of the result and analysis depicts, training cascade and distributed SVM’s algorithms for incremental large scale dataset and fine tuning SVM hyper-parameter in distributed environment using parallel methods performs better comparatively traditional algorithms.