An Investigation of Machine Learning Techniques for Effective Intrusion Detection

Yingzhi Yang, Shengjun Zhai, Jiangnan Fu, Yuchan Kim and Shulin Ji

With technological advancement, cybercrime is increasing exponentially. Cyber resources are highly vulnerable to intrusions, so they cannot be fully protected from such threats through conventional cybersecurity methods such as the firewall. Hence, there is an urgent need for more sophisticated and specialised systems for detecting unwanted intrusions. Recently, machine learning (ML) techniques have gained close attention from security experts based on their advantages such as adaptability, flexibility, and learning by example. This paper investigates various ML techniques for effective intrusion detection by comparing their performance using the KDD benchmark dataset for popular performance metrics. The results indicate that CART is a more accurate ML technique, detecting upto 99.7% of all intrusions, whereas ZeroR is the faster technique for building a model from the KDD dataset. A comparative analysis of these techniques identifies the best-performing ML technique, which can be a candidate for developing an effective intrusion detection system. The results provide a better understanding of the framework for applying ML techniques to intrusion detection in particular and any classification task in general.

Volume 11 | Issue 1

Pages: 22-30