Pengaruh Sampling Method dan Feature Extraction untuk Meningkatkan Detection Rate pada Minority Class pada Intrusion Detection System yang Disusun dari Support Vector Machine, Decision Tree, dan Naïve Bayes

Janthake Decuellar (Program Studi Informatika)
Henry Novianus Palit (Program Studi Informatika)
Justinus Andjarwirawan (Program Studi Informatika)

Jurnal Infra
Vol 9, No 2 (2021)

Intrusion detection system (IDS) has started to rely on machine learning to perform misuse detection or anomaly detection. As misuse detection, machine learning must be able to detect various types of intrusions, including those that are rare. However, machine learning has weaknesses, especially when faced with imbalanced datasets. Various methods are used to make machine learning able to perform the classification correctly even though the data provided is imbalanced. One of them in this study tries to implement Principal Component Analysis as feature extraction, Tomek Links as under-sampling and ADASYN as over-sampling on datasets. There are two types of datasets used in this research, namely KDD-99 and UNSW-NB15.The results obtained from research on the KDD’99 dataset are, Support Vector Machine can identify more intrusions than before and True Positive Rate of Decision Tree model for minority classes is increased between 0.03% to 4.762%. The results obtained from research on UNSW-NB15 dataset, accuracies for Support Vector Machine and Naïve Bayes models are increased between 0.045% to 1.513%.