ميترا اميدوار

Software Defined Network (SDN) offers significant advantages in network management by separating the control plane from the data plane. This architecture enables centralized control, dynamic resource allocation, an‎d flexible configuration. However, it creates new security challenges that the integration of an Intrusion Detection System (IDS) into this architecture can increase network security. In recent years, the use of Deep Learning (DL) has made it possible to process large amounts of data as well as detect anomalies more accurately. In addition, the use of Feature selec‎tion (FS) methods have improved the performance of intrusion detection systems due to the problem of overfitting an‎d false positive rates. In this thesis, using the combination of FS an‎d DL methods, we present an accurate IDS that maintains a good balance between precision an‎d recall criteria. Using a hybrid FS method based on Recursive Feature Elimination (RFE) combined with L1 an‎d L2 regularization techniques an‎d Tree-based methods, Decision tree (DT), Ran‎dom Forest (RF), an‎d XGBoost, the important features of the dataset are selec‎ted, an‎d then using Convolutional Neural Network (CNN), we perform the classification. The results of multiple eva‎luations show that Tree-based methods have achieved better results than regularization techniques in eva‎luation criteria. The combined RF+CNN method achieved an F1-score of more than 99.80% on the InSDN dataset.

شبكه‌هاي مبتني‌بر نرم‌افزار , سامانه‌هاي تشخيص نفوذ , انتخاب ويژگي , يادگيري عميق

Software Defined Network , Intrusion detection system , Feature selec‎tion , Deep Learning

Mitra Omidvar

Dr. Shahhoseini