Adaptive Anomaly Detection Framework Model Objects in Cyberspace.

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Rapid growth in telecommunications necessitates robust network security solutions.
• E-banking, e-commerce, and business data sharing face significant threats from cyberattacks.
• Current network monitoring and security are complex challenges for administrators.

Purpose of the Study:

• To propose a methodology for high-level, sustainable protection against cyberattacks.
• To develop an adaptive anomaly detection framework using deep and machine learning.
• To improve cybersecurity systems by effectively classifying network intrusions.

Main Methods:

• Developed an adaptive anomaly detection framework utilizing deep learning (LSTM-RNN) and machine learning (SVM, K-NN) algorithms.
• Employed the information gain method for relevant feature selection from network datasets.
• Evaluated the model on standard datasets: KDD Cup'99, NSL-KDD, ISCX, and ICI-ID2017 for classifying DoS and DDoS attacks.

Main Results:

• The Long-Short Term Memory Recurrent Neural Network (LSTM-RNN) algorithm achieved the highest accuracy in detecting cyber threats.
• The proposed LSTM-RNN based system demonstrated exceptional testing accuracy rates of 99.51% and 99.91% on the evaluated datasets.
• Comparative analysis confirmed the superiority of LSTM-RNN over SVM and K-NN for anomaly-based cybersecurity detection.

Conclusions:

• The LSTM-RNN model is highly efficient and effective for enhancing cybersecurity systems.
• The proposed framework provides a sustainable and high level of protection against cyberattacks.
• Anomaly detection using deep learning, specifically LSTM-RNN, is crucial for modern network security.