Abstract: Network traffic classification is a process of finding type of end user applications toward network planning and bandwidth management, diagnostic monitoring, traffic analysis, prediction and engineering, anomalous traffic detection and QoS provisioning. Today with the improvement in field of information security, traditional network traffic classification such as payload based and port based classification are useless. Supervised, unsupervised and semi-supervised are three machine learning algorithms suggested to tackle traditional techniques. In this paper a semi-supervised approach including clustering (EM clustering, DBSCAN and k-Means), mapping and J48 classification is proposed assuming random 20, 50 and 80 percent of NSL-KDD dataset as unlabeled class attributes. Weka 3.7.11 is used for this implementation and overall precision, recall and F-Measure are the metric of performance evaluation comparing the results with 100 percent labeled training dataset. The results showed that the overall recall, precision and F-Measure of 20 and 80 percent of unlabeled dataset are more than 95.6% and for 50 percent unknown traffic flows is 90.1%. This measurement for full labeled training dataset is 98.8%.
Keywords: Traffic classification, Machine Learning, EM clustering, DBSCAN, K-Means, J48 classification, NSL-KDD dataset.
Title: IMPLEMENTING A NEW SEMI-SUPERVISED APPROACH FOR INTERNET TRAFFIC CLASSIFICATION USING NSL-KDD DATASET
Author: Vajihe Abdi, Marzieh AhmadZadeh
International Journal of Computer Science and Information Technology Research
ISSN 2348-120X (online), ISSN 2348-1196 (print)
Research Publish Journals
