Deep Learning Method for Denial of Service Attack Detection Based on Restricted Boltzmann Machine.

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Denial of Service (DoS) attacks pose a significant threat to network availability.
• Existing deep learning methods for DoS detection require further accuracy improvements.

Purpose of the Study:

• To propose and evaluate a novel deep learning model for accurate DoS attack detection.
• To enhance the performance of Restricted Boltzmann Machines (RBMs) for network security applications.

Main Methods:

• Application of a deep learning method based on a Gaussian-Bernoulli type Restricted Boltzmann Machine (RBM).
• Incorporation of seven additional layers between visible and hidden layers to create a deep RBM architecture.
• Optimization of hyperparameters for the proposed deep RBM model.
• Utilizing a variant of RBM suitable for continuous data, featuring a Gaussian distribution for the visible layer.

Main Results:

• The proposed multilayer deep Gaussian-Bernoulli RBM achieved higher detection accuracy for DoS attacks.
• Comparative analysis demonstrated superior performance over standard Bernoulli-Bernoulli RBM, Gaussian-Bernoulli RBM, and deep belief network methods.
• The model's effectiveness was validated using the NSL-KDD dataset.

Conclusions:

• The developed deep Gaussian-Bernoulli RBM offers a more accurate and effective approach to DoS attack detection.
• The multilayer architecture and continuous data handling significantly boost detection capabilities.
• This method presents a promising advancement in network intrusion detection systems.