Search for collections on Universitas Indo Global Mandiri Repository

EVALUASI PERFORMA XGBOOST MENGGUNAKAN TEKNIK FEATURE SELECTION UNTUK SISTEM INTRUSION DETECTION

At'thoriq, Salsabiil and Gustriansyah, Rendra and Irfani, Muhammad Haviz (2026) EVALUASI PERFORMA XGBOOST MENGGUNAKAN TEKNIK FEATURE SELECTION UNTUK SISTEM INTRUSION DETECTION. Masters thesis, Universitas Indo GLobal Mandiri.

[thumbnail of 2022110148_JUL_2026_1.pdf] Text
2022110148_JUL_2026_1.pdf
Restricted to Repository staff only

Download (13MB) | Request a copy
[thumbnail of Salsabiil At'thoriq_2022110148_File Full Karya Ilmiah.pdf] Text
Salsabiil At'thoriq_2022110148_File Full Karya Ilmiah.pdf
Restricted to Repository staff only

Download (2MB) | Request a copy
[thumbnail of Salsabiil At'thoriq_2022110148_File Cover-Daftar isi.pdf]
Preview
Text
Salsabiil At'thoriq_2022110148_File Cover-Daftar isi.pdf

Download (604kB) | Preview

Abstract

Perkembangan ancaman siber yang semakin kompleks menuntut sistem keamanan jaringan memiliki kemampuan deteksi intrusi yang akurat dan efisien. Salah satu pendekatan yang banyak digunakan adalah Intrusion Detection System (IDS) berbasis machine learning. Namun, tingginya dimensi fitur pada dataset jaringan modern seperti UNSW-NB15 dapat menurunkan performa model akibat adanya fitur yang tidak relevan atau redundan. Penelitian ini bertujuan untuk mengevaluasi performa algoritma XGboost dengan menerapkan teknik Feature selectionndalam sistem intrusion detection. Metode Feature selectionnyang digunakan adalah kombinasi Chi-Square sebagai metode filter untuk seleksi awal fitur dan Feature importance XGboost sebagai metode embedded untuk seleksi lanjutan. Dataset yang digunakan dalam penelitian ini adalah UNSW-NB15 dengan skenario pembagian data Stratified Hold-Out Validation pada rasio 60:40, 70:30, 80:20 dan 90:10. Hasil penelitian menunjukkan bahwa Model 1 (XGboost tanpa Feature selection) menghasilkan akurasi tertinggi sebesar 97,69% menggunakan 42 fitur. Sementara itu, Model 3 yang menggunakan kombinasi Chi-Square dan Feature importance XGboost mampu mencapai performa yang sangat kompetitif dengan akurasi 97,44% hanya dengan menggunakan 10 fitur. Penerapan teknik Feature selectionnkombinasi ini terbukti efektif meningkatkan efisiensi model dengan mereduksi jumlah fitur sebesar 76,19% tanpa menyebabkan penurunan performa yang signifikan, sehingga menghasilkan keseimbangan optimal antara performa klasifikasi dan kompleksitas komputasi pada sistem IDS.
Kata Kunci: Intrusion Detection System, XGboost, Feature selection, Chi-Square, UNSW-NB15, Machine learning.

The growing complexity of cyber threats demands that network security systems possess accurate and efficient intrusion detection capabilities. One widely adopted approach is the implementation of machine learning-based Intrusion Detection Systems (IDS). However, the high dimensionality of features in modern network datasets, such as UNSW-NB15, can degrade model performance due to irrelevant or redundant attributes. This study aims to evaluate the performance of the XGboost algorithm by applying Feature selectionntechniques within an Intrusion Detection System. The proposed Feature selectionnmethod employs a combination of Chi-Square as a filter method for initial Feature selectionnand XGboost Feature importance as an embedded method for further refinement. The dataset used in this study is UNSW-NB15, evaluated using a Stratified Hold-Out Validation split at an 80:20 ratio. The empirical results demonstrate that Model 1 (XGboost without feature selection) achieves the highest absolute Accuracy of 97.69% using 42 features. Meanwhile, Model 3, which utilizes the combined Chi-Square and XGboost Feature importance method, delivers a highly competitive Accuracy of 97.44% using only 10 features. The application of this hybrid Feature selectionntechnique successfully reduces the feature dimension by 76.19% without a significant drop in classification performance, thereby achieving an optimal balance between Accuracy and computational efficiency for modern IDS.
Keywords: Intrusion Detection System, XGboost, Feature selection, Chi-Square, UNSW-NB15, Machine learning.

Item Type: Thesis (Masters)
Subjects: Q Science > Q Science (General)
Divisions: Fakultas Ilmu Komputer dan Sains > Teknik Informatika S1
Depositing User: salsabiil abel at'thoriq
Date Deposited: 12 Aug 2026 01:09
Last Modified: 12 Aug 2026 01:09
URI: https://repository.uigm.ac.id/id/eprint/6962

Actions (login required)

View Item
View Item