Network Intrusion Detection Technology Based on Convolutional Neural Network and BiGRU.

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Existing network intrusion detection systems suffer from low accuracy and high false-alarm rates, particularly in multi-class intrusion scenarios.
• Network traffic data often presents challenges with high dimensionality and imbalanced sample distributions, hindering model performance.

Purpose of the Study:

• To propose a novel network intrusion detection model that enhances accuracy and reduces false alarms for multiple intrusion classifications.
• To address feature dimensionality and sample imbalance issues in network traffic data.

Main Methods:

• A hybrid sampling algorithm combining ADASYN and RENN was employed for data preprocessing.
• Feature selection was performed using a combination of the random forest algorithm and Pearson correlation analysis.
• A convolutional neural network (CNN) was utilized for spatial feature extraction, followed by average and max pooling.
• Bidirectional Gated Recurrent Unit (BiGRU) was incorporated for extracting long-distance dependent information features.
• The Softmax function was used for final classification.

Main Results:

• The proposed model achieved high accuracy rates of 85.55% on UNSW_NB15, 99.81% on NSL-KDD, and 99.70% on CIC-IDS2017 datasets.
• The model demonstrated superior performance compared to the CNN-GRU model, with improvements of 1.25%, 0.59%, and 0.27% on the respective datasets.
• Effective feature learning was achieved through the combined CNN and BiGRU architecture.

Conclusions:

• The proposed CNN-BiGRU network intrusion detection model effectively addresses challenges of low accuracy and high false-alarm rates.
• The integration of hybrid sampling, feature selection, and a hybrid deep learning architecture leads to comprehensive and robust intrusion detection.
• The model shows significant potential for real-world application in network security.