Akademik Veri Yönetim Sistemi
JOURNAL OF THE FACULTY OF ENGINEERING AND ARCHITECTURE OF GAZI UNIVERSITY, cilt.39, sa.2, ss.679-692, 2023 (SCI-Expanded)
Purpose:
The aim of the study is to increase the performance (speed and accuracy) of intrusion detection models
developed with machine learning methods.
Theory and Methods:
In this study; A four-stage methodology has been proposed to examine the effect of preprocesses applied to
the data sets on the success of intrusion detection models developed using machine learning models. Firstly,
preprocesses such as the categorical data encoding, scaling and hybrid feature selection were applied to the
data sets, then, the intrusion detection models were developed with preprocessed datasets and machine
learning algorithms, and, finally, the performance of the models was evaluated and improved by performing
hyper-parameter optimization.
Results:
Experimental results showed that the proposed methodology increased the performance (speed and accuracy)
of the intrusion detection models, namely, the accuracy of 96.1% and the speed of 0.373 second were
obtained in the training data set and the accuracy of 100% and the speed of 0.005 second were obtained in
the test dataset.
Conclusion:
In this study; Experimental results showed that preprocessing and hyper-parameter optimization in models
increase the intrusion detection speed and accuracy. The most successful results were obtained with the
datasets in which all preprocesses were applied together. It has been concluded that in order to achieve more
successful results the up-to-date and different data sets should be used, different methods should be tried for
each algorithm used, and hyper-parameter optimization should be done.