Exploring A Two-Phase Deep Learning Framework For Network Intrusion Detection
Ragam Padmaja, Papa Rao Challagundla
Abstract
Intrusion detection systems (IDS) often use machine learning and deep learning to identify and categorize cyber-attacks swiftly. However, as these attacks expand, a robust response becomes crucial. Despite the availability of numerous intrusion detection datasets, there’s limited research assessing model performance across various public datasets. Continuous updates and benchmarking of these datasets remain crucial due to evolving attack strategies. This study compares the proposed model with convolutional neural network (CNN) and deep neural network (DNN) models to create a versatile and effective IDS. Adapting IDS to handle evolving network behaviours and increasing assaults necessitates dynamic approaches with large datasets. Combining autoencoders (AE) and Long Short-Term Memory (LSTM) in our new two-stage deep learning method demonstrates effectiveness in detecting assaults using CICIDS2017 and CSE-CICDIS2018 datasets.