A Hybrid Spectral Clustering and Deep Neural Network Ensemble Algorithm for Intrusion Detection in Sensor Networks

Area of Science:

• Cybersecurity
• Network Security
• Machine Learning

Background:

• Intrusion detection systems (IDS) are crucial for network defense.
• Detecting disguised malicious traffic poses a significant challenge.
• Existing IDS methods struggle with sophisticated network attacks.

Purpose of the Study:

• To propose a novel intrusion detection system (IDS) named SCDNN.
• To enhance the detection of malicious network traffic disguised as normal access.
• To improve the accuracy and scope of intrusion detection in large networks.

Main Methods:

• Combines spectral clustering (SC) for data partitioning with deep neural networks (DNN) for classification.
• Divides datasets into subsets based on sample similarity using SC.
• Measures similarity features between training and testing data for DNN analysis.

Main Results:

• SCDNN demonstrated superior performance compared to Backpropagation Neural Network (BPNN), Support Vector Machine (SVM), Random Forest (RF), and Bayes Tree models.
• Achieved higher detection accuracy and identified a broader range of abnormal attack types.
• Validated on KDD-Cup99, NSL-KDD, and a sensor network dataset.

Conclusions:

• SCDNN offers an effective approach for intrusion detection in large-scale networks.
• The hybrid SC-DNN model significantly improves upon traditional machine learning techniques for IDS.
• This method provides a valuable tool for network security analysis and defense.