Detection of Network Anomalies with Machine Learning Methods

The present study, aimed to detect cyber-attacks, and unexpected access requests on devices in the telecommunication networks, enabling the necessary measures to be taken early. With K-Nearest Neighbors (KNN) and Naive Bayes machine learning methods, predicted whether the raw data packets contain cyber-attack according to different properties of these packets using the UNSW-NB15 dataset. KNN algorithms with different K values and the Naive Bayes method were compared according to accuracy rates and the results were given in the table. As a result, changes in accuracy rates were observed according to different k neighbor values in the KNN algorithm. Higher accuracy rates than Naive Bayes were achieved in the models created with the KNN algorithm.

I. Introduction

Along with the benefits, information technologies also bring some important threats and risks. Cyber-attacks, one of the most important of these threats, are crucial for businesses and personal security. At the same time, it is seen that cybercrime causes very serious damage and costs worldwide [1].