Measuring throughput and latency of machine learning techniques for intrusion detection

Winfred Yaokumah, Charity Y.M. Baidoo, Ebenezer Owusu

Research output: Chapter in Book/Report/Conference proceedingChapterpeer-review

Abstract

When evaluating the effectiveness of machine learning algorithms for intrusion detection, it is insufficient to only focus on their performance metrics. One must also focus on the overhead metrics of the models. In this study, the performance accuracy, latency, and throughput of seven supervised machine learning algorithms and a proposed ensemble model were measured. The study performs a series of experiments using two recent datasets, and two filter-based feature selection methods were employed. The results show that, on average, the naive bayes achieved the lowest latency, highest throughput, and lowest accuracy on both datasets. The logistics regression had the maximum throughput. The proposed ensemble method recorded the highest latency for both feature selection methods. Overall, the Spearman feature selection technique increased throughput for almost all the models, whereas the Pearson feature selection approach maximized performance accuracies for both datasets.

Original languageEnglish
Title of host publicationHandbook of Research on AI and ML for Intelligent Machines and Systems
PublisherIGI Global
Pages27-49
Number of pages23
ISBN (Electronic)9798369300008
ISBN (Print)1668499991, 9781668499993
DOIs
Publication statusPublished - 27 Nov 2023

Fingerprint

Dive into the research topics of 'Measuring throughput and latency of machine learning techniques for intrusion detection'. Together they form a unique fingerprint.

Cite this