Akademik Veri Yönetim Sistemi
16th International Conference on Mobile Web and Intelligent Information Systems, MobiWIS 2019, İstanbul, Türkiye, 26 - 28 Ağustos 2019, cilt.11673 LNCS, ss.111-122, (Tam Metin Bildiri)
Since three decades, artificial intelligence has been evolved in order to outperform the tasks that human beings are not capable. These tasks can be any problem from our lives and one of these problems is computer networks-related tasks which huge number of privacy data is transferred even a second. Within last two decades, machine learning techniques with capabilities for prediction, optimisation, and as well as classification are developed for using to solve the real-life problems. In this paper, challenging and popular NSL-KDD dataset for intrusion detection is chosen for performed experiments, where classification and three benchmark machine learning techniques are used in order to determine optimum technique for classification domain. Experiments are performed by implementing 3-layered Back-propagation Neural Network, Support Vector Machine and Decision Tree. Thirty percent (30%) of instances of NSL-KDD Dataset were considered that causes 25193 of total instances in experiments. Each experiment is repeated for two times by using 60% and 70% of instances for training and the rest for testing. Increment of training patterns or instances caused little fluctuations on accuracy rates in Decision Tree and Back-propagation but it causes more effect in Support Vector Machine which is about 1% decrement in accuracy rate. It is seen from the performed experiments’ results that, increment or degradation of training ratio of instances in dataset does not affect the performance of the techniques directly.