Comparison of Data Mining Algorithm: PSO-KNN, PSO-RF, and PSO-DT to Measure Attack Detection Accuracy Levels on Intrusion Detection System

Abstract

Nowadays, computer networks are widely used to exchange valuable and confidential data information between servers to computers or cellular devices. Access to user control and use of software or hardware as a firewall often experience security problems. Unauthorized access to information through computer networks continues to occur and tends to increase. This study examines the attack detection mechanism by using three data mining algorithms based on particle swarm optimization (PSO), namely PSO-K Nearest Neighbor, PSO-Random Forest, and PSO-Decision Tree in the Canadian Institute for Cybersecurity Dataset (CICIDS2017). The initial experiment showed that the approach using the PSO-RF method was able to produce the highest accuracy of attack detection. Accuracy values generated using the PSO-RF algorithm with a combination of the number of trees and maximal depth = 20 in the CICIDS2017 dataset are intact higher than other proposed algorithms. The highest accuracy of attack detection in the CICIDS2017 dataset is intact, which is 99.76%. In the CICIDS2017 dataset 50% Benign and 50% Attack it turns out that the PSO-RF algorithm with a combination of the number of trees and maximal depth = 20 also gets the highest accuracy value of 99.67%.