부산대학교 도서관

Traffic classification is very important for network events management, such as data transverse to security monitoring, QOS, and accounting, to providing Internet service providers with beneficial insights for services provisioning. Yet, traffic classification accuracy is a strenuous topic because information known to the network admin, i.e., Port number, packet-headers, which does not contain sufficient forecast to consent for an accurate and efficient methodology. This leads to conventional techniques for flow classification that are often no further accurate than 35-65%. Machine learning algorithms are presented to solve complex, high-level abstract and random data, the nearest neighbor (NN)-based technique has exhibited excellent classification performance. It also has numerous major advantages, such as no necessities of training practice, no risk of overfitting of attributes, and logically being able to handle a vast number of classes. Deep learning architectures algorithms are convolutional Neural Network (CNN), K-nearest neighbors (KNN), Recurrent Neural Networks (RNNs), Support vector machines (SVMs), deep neural network (DNN), Decision tree C4.5, Naive bayes, Long Short-Term Memory Networks (LSTMs), Linear Discriminant Analysis (LDA) Nevertheless, In this study, we propose some techniques of improving traffic classification accuracy either by combining two or more algorithms through a multi-level classifier or by implementing computer aided techniques or using ensemble methods to obtain better traffic classification accuracy.