Banca de DEFESA: Moises Silva de Sousa

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : Moises Silva de Sousa
DATE: 20/12/2023
TIME: 08:30
LOCAL: https://teams.microsoft.com/l/meetup-join/19%3a0a9998084fcc459f92b72fb6608767f7%40thread.tacv2/17030
TITLE:

The use of Feature Engineering to optimize the performance of supervised machine learning models applied to Intrusion Detection Systems.

KEY WORDS:

Cybersecurity, Feature Engineering, CFS Subset, Information Gain, Correlation, SVM, artificial intelligence, network anomaly.

PAGES: 106
BIG AREA: Engenharias
AREA: Engenharia Elétrica
SUMMARY:

The use of machine learning (ML) techniques for building intrusion detection systems (IDS) has been growing every year. Numerous ML technologies have been emerged allowing to build predictive learning models in order to identify and detect network traffic anomalies using IDS. A part of the ML techniques is a nonparameterized approach, extracting data from large datasets in an undiscriminated way which includes irrelevant and redundant data, affecting adversely the performance of the ML classification algorithms. However, it is possible to provide to a ML technique the ability to properly extract data from the dataset by selecting an appropriate subset of attributes, i.e., by means of feature engineering (FE), that allows to improve the performance of the data extraction, training and classification ML processes. This work discusses how feature engineering can be used to improve the ML processes in IDS systems. In particular, it demonstrates that with an appropriate selection of attributes, the training process can be disrupted, improving the processing speed while maintaining the desired classification accuracy. The performance evaluation experiments are based on the WEKA software platform using the dataset NSL-KDD and the Support-Vector Machine (SVM) as machine learning classification algorithm. By using different data testtraining division ratios (60- 40, 70- 30 and 80-20) and attribute selection techniques (Information Gain, Correlation Gain and Correlation-based Feature Selection – CFS) this work achieves results that allow to understand how feature engineering may impact positively the performance of an ML-IDS system.

COMMITTEE MEMBERS:
Interno - 2311780 - FABIO LUCIO LOPES DE MENDONCA
Interno - 2556078 - GEORGES DANIEL AMVAME NZE
Externo à Instituição - LEANDRO ALVES NEVES - UNESP
Presidente - 330495 - WILLIAM FERREIRA GIOZZA