A hybrid feature learning model to enhance multilayer perceptron for network intrusion detection

Authors

  • NOOR ABDULKADHIM وزارة التربية العراقية
  • Oras Nasef Jasim General Directorate of Education in Thi-Qar Governorate, Thi-Qar, Iraq
  • Zainab Kashlan Yaser General Directorate of Education in Thi-Qar Governorate, Thi-Qar, Iraq

DOI:

https://doi.org/10.32792/jeps.v15i1.505

Keywords:

Classification, Class imbalance Random forest

Abstract

Abstract

Technologies of computer and network have an essential role in our daily life, although, the advantages of them have been balanced with important attacks of network in the last few years. Intrusion detection of network has been increasingly adopted as the efficient technique for coping with threats of network. For classifying network intrusion, techniques based on noninvasive like machine learning are effective and reliable. However, a lot of supervised and unsupervised learning methods exist from a domain of machine learning which are utilized for increasing prediction network intrusion accuracy, present algorithms of detection as well as techniques of improvement utilized are the issue for obtaining good performance. Many extra and non-related data in high-dimensional sets of data interfere with intrusion detection system (IDS) classification process.  Non-related features in data influence model accuracy also raise time of training required for creating a model. Feature selection refers to the basic for creating IDS. For improving detection performance, we presented: At first, the balanced set of data can be

launched by utilizing both methods of oversample and undersampling. Second, a technique of dimensionality reduction is according to the selection and extraction of features. Third, utilizing algorithms of deep learning for detecting network intrusion. At last, the proposed model performance is validated by applying a dataset of NSLKDD. Experimental results show our proposed work achieves a high level of accuracy in the prediction of network intrusion.

 

Downloads

Published

2025-03-01