REDUKSI DIMENSI DATA MENGGUNAKAN KOMBINASI METODE SELEKSI FITUR FILTER VARIAN RENDAH DAN RELIEFF DALAM PREDIKSI SERANGAN BOTNET

MURYANTO, NUR DWI (2021) REDUKSI DIMENSI DATA MENGGUNAKAN KOMBINASI METODE SELEKSI FITUR FILTER VARIAN RENDAH DAN RELIEFF DALAM PREDIKSI SERANGAN BOTNET. S2 thesis, Universitas Mercu Buana Jakarta.

Preview	Text (HAL COVER) Cover - Thesis Nur Dwi Muryanto - 55419110012.pdf Download (397kB) | Preview
	Text (BAB I) Bab I - Thesis Nur Dwi Muryanto - 55419110012.pdf Restricted to Registered users only Download (690kB)
	Text (BAB II) Bab II - Thesis Nur Dwi Muryanto - 55419110012.pdf Restricted to Registered users only Download (306kB)
	Text (BAB III) Bab III - Thesis Nur Dwi Muryanto - 55419110012.pdf Restricted to Registered users only Download (259kB)
	Text (BAB IV) Bab IV - Thesis Nur Dwi Muryanto - 55419110012.pdf Restricted to Registered users only Download (992kB)

Abstract

Botnet is one kind of malware that become serious threat in cyber security, especially in IoT devices. Therefore, it is necessary to predict a botnet attack. One of the techniques used is Machine learning-based, which is based on data (knowledge) to be studied and made a mathematical model. However, a common problem in predicting attacks in the era of big data is the large dimension of the Dataset which affects its computational speed, while it is possible that not all features are relevant in attack prediction. Therefore, in this study proposed feature reduction in the Dataset will using a combination of the Low Variant Filter feature selection method and ReliefF on the machine learning algorithm Support Vector Machine. Therefore, in this study, a feature reduction method in the Dataset is proposed using a combination of two feature selection methods, namely Low Variance Filter and ReliefF, on the Support Vector Machine. When the Low Variance Filter was performed in front of ReliefF (Scenario 2), there was an increase in computational speed with an average value of 39.1 times faster than without using feature selection (Scenario 1). Meanwhile, if ReliefF is applied before the Low Variant Filter (Scenario 3), it reaches an average of 30.91 times faster. In addition, the model evaluation shows an increase in Accuracy, Precision, and Specificity, namely an increase in accuracy, with an average of 38.627% (Scenario 2) and 38.059% (Scenario 3); increased precision, with an average of 42.839% (Scenario 3); and 42.452% (Scenario 3), as well as an increase in Specificity with an average of 81.14% (Scenario 2) and 84.336% (Scenario 3). However, there was a decrease in recall although it was not significant with an average of 4.453% (Scenario 2) and 5.844% (Scenario 3). Keywords: Botnet, SVM, Feature Selection, Low Variance Filter, ReliefF Botnet merupakan salah satu jenis malware yang menjadi ancaman serius dalam dunia keamanan siber, terutama dalam perangkat IoT. Oleh karena itu, diperlukan upaya prediksi terhadap suatu serangan Botnet. Salah satu teknik yang digunakan adalah Machine learning-based, yang berbasis data (pengetahuan) untuk dipelajari dan dibuat suatu model matematis. Namun, permasalahan umum dalam prediksi serangan di era big data adalah dimensi Dataset yang besar. Hal ini berpengaruh pada kecepatan komputasinya, sementara itu ada kemungkinan tidak semua fitur relevan dalam prediksi serangan. Oleh karena itu, dalam penelitian ini akan dilakukan reduksi fitur dalam Dataset menggunakan kombinasi metode seleksi fitur Filter Varian Rendah dan ReliefF pada algoritma machine learning Support Vector Machine. Saat Filter Varian Rendah dilakukan di depan ReliefF (Skenario 2), terjadi peningkatan kecepatan komputasi dengan nilai rata-rata mencapai 39,1 kali lebih cepat dibandingkan tanpa menggunakan seleksi fitur (Skenario 1). Sedangkan jika diterapkan ReliefF sebelum Filter Varian Rendah (Skenario 3) mencapai ratarata 30,91 kali lebih cepat. Selain itu, evaluasi model menunjukkan peningkatan Akurasi, Presisi, dan Specificity, yaitu peningkatan Akurasi, dengan rata-rata 38,627% (Skenario 2) dan 38,059% (Skenario 3); peningkatan Presisi, dengan ratarata 42,839% (Skenario 3); dan 42,452% (Skenario 3), serta peningkatan Specificity dengan rata-rata 81,14% (Skenario 2) dan 84,336% (Skenario 3). Namun terjadi penurunan Recall meskipun tidak signifikan dengan rata-rata sebesar 4,453% (Skenario 2) dan 5,844% (Skenario 3). Kata Kunci: Botnet, SVM, Seleksi Fitur, Filter Varian Rendah, ReliefF

Actions (login required)

View Item