Research on intrusion detection model based on improved MLP algorithm.

Area of Science:

• Cybersecurity
• Machine Learning
• Network Intrusion Detection

Background:

• Imbalanced datasets in intrusion detection systems (IDS) compromise accuracy, risking undetected malicious traffic and financial losses.
• Multi-layer perceptrons (MLP) are suitable for intrusion detection due to their ability to model complex, nonlinear attack patterns with activation functions like ReLU and Sigmoid.
• Standard MLPs have limited feature extraction, hindering minority class classification in imbalanced datasets.

Purpose of the Study:

• To improve the recognition of minority classes in imbalanced datasets for intrusion detection.
• To enhance the feature extraction capabilities of Multi-layer Perceptrons (MLP) for more accurate threat identification.
• To develop a novel algorithm integrating Convolutional Neural Networks (CNNs) and attention mechanisms with MLP.

Main Methods:

• Integration of AlexNet's feature extraction module with MLP architecture.
• Incorporation of the SKNet attention mechanism to focus on critical minority class features.
• Utilizing ReLU and Sigmoid activation functions within the MLP for nonlinear transformation.

Main Results:

• The enhanced MLP algorithm demonstrated superior performance over standard MLP in all seven classification tasks.
• Significant F1 score improvements were observed: 18.93% for BotnetARES and 26.57% for PortScan.
• The proposed method effectively addresses the challenge of classifying minority classes in imbalanced intrusion detection datasets.

Conclusions:

• The integration of AlexNet's feature extraction and SKNet attention mechanism significantly boosts MLP's performance in imbalanced intrusion detection.
• The enhanced algorithm offers a promising solution for detecting sophisticated and low-frequency cyber threats.
• This approach validates the efficacy of combining deep learning feature extraction with attention mechanisms for robust intrusion detection.