Network Anomaly Intrusion Detection Based on Deep Learning Approach.

Area of Science:

• Cybersecurity
• Machine Learning
• Network Security

Background:

• Internet traffic analysis is crucial for detecting cyberattacks.
• Deep learning shows promise for intrusion detection systems (IDS).
• Existing research often uses outdated datasets, limiting the detection of current threats.

Purpose of the Study:

• To evaluate the effectiveness of various deep learning models for network intrusion detection.
• To assess model performance on a current dataset (CSE-CIC-IDS2018).
• To compare individual models against combined models for practical IDS implementation.

Main Methods:

• Preprocessing the CSE-CIC-IDS2018 dataset.
• Constructing and evaluating six deep learning models: DNN, CNN, RNN, LSTM, CNN + RNN, and CNN + LSTM.
• Conducting binary and multi-class classification experiments to identify benign and malicious traffic (BruteForce, DoS, Web Attacks, Infiltration, Botnet, DDoS).

Main Results:

• All evaluated models achieved high accuracy, with multi-class classification exceeding 98%.
• Individual models (DNN, RNN, CNN) demonstrated competitive detection performance.
• Combined models (CNN + RNN, CNN + LSTM) exhibited longer inference times compared to individual models.

Conclusions:

• Deep learning models, particularly DNN, RNN, and CNN, are effective for network intrusion detection.
• Individual models offer a better balance of performance and efficiency for IDS implementation.
• The study highlights the importance of using current datasets for evaluating intrusion detection mechanisms.