University of Maryland Global Campus, Adelphi, USA
*Corresponding Author: Balakrishnan Dasarathy, University of Maryland Global Campus, Adelphi, USA.
Received: May 01, 2021; Published: June 09, 2021
This paper is on applying leading machine learning (ML) techniques to detect cyberthreats and cyberattacks. The paper begins with a short overview of three leading supervisory techniques, Logistic Regression, Neural Network (NN) and Support Vector Machine, and one unsupervised learning technique, the Multivariate Gaussian Distribution-based anomaly detection, that are being applied. The datasets used for training, validation and testing are then described. Potential alternatives to the datasets used are also discussed. Metrics used to assess the algorithms, precision, recall and accuracy, are then defined. An overview of bias vs. variance tradeoff to get optimal results with the algorithms is then provided. The performance of the ML algorithms applied is then described. The performance of the NN algorithm for two-class classification (normal vs. attack) and the anomaly detection using the Multivariate Gaussian Distribution function is encouraging. The results are summarized, and future directions are outlined. A main goal of the future research is to improve performance in classifying attacks into their constituent classes, i.e., to develop more accurate signatures or models for attack classes. An in-depth bias vs. variance analysis of the algorithms applied is also a major direction going forward.
Keywords: Intrusion Detection; Machine Learning; Cyberattacks; Cyberthreats; Machine Learning for Cyberthreat and Cyberattack Detection
Citation: Balakrishnan Dasarathy. “A Comparative Study of Machine Learning Techniques in Cyberthreat and Cyberattack Detections1". Acta Scientific Computer Sciences 3.7 (2021): 29-40.
Copyright: © 2021 Balakrishnan Dasarathy. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.