A Lightweight Intelligent Network Intrusion Detection System Using One-Class Autoencoder and Ensemble Learning for...

Area of Science:

• Cybersecurity
• Machine Learning
• Internet of Things (IoT)

Background:

• Traditional network intrusion detection systems struggle with unknown (zero-day) attacks.
• Retraining models for new threats is slow and requires expert intervention.
• Existing systems lack efficiency in identifying novel attack patterns in IoT environments.

Purpose of the Study:

• To develop a lightweight, intelligent network intrusion detection system (NIDS) for enhanced IoT cybersecurity.
• To accurately identify normal, abnormal, and unknown network traffic.
• To improve the detection rate of novel cyber threats without constant retraining.

Main Methods:

• Implementation of a One-Class Bidirectional GRU Autoencoder model trained on normal network data.
• Application of ensemble learning with Soft Voting for robust classification of network traffic.
• Validation using benchmark datasets: WSN-DS, UNSW-NB15, and KDD CUP99.

Main Results:

• Achieved high recognition rates: 97.91% on WSN-DS, 98.92% on UNSW-NB15, and 98.23% on KDD CUP99.
• The proposed model accurately distinguishes normal from abnormal data, including unknown attacks.
• Effectively classifies unknown attacks by identifying the most similar known attack type.

Conclusions:

• The proposed lightweight intelligent NIDS demonstrates high feasibility and efficiency for IoT cybersecurity.
• The combination of One-Class Bidirectional GRU Autoencoder and ensemble learning offers superior performance in detecting known and unknown threats.
• The algorithm's portability across different datasets confirms its practical applicability.