Selective Attention-based Temporal Learning for Network Security Status Prediction

Selective Attention-based Temporal Learning for Network Security Status Prediction

연구 분야: Cryptography

학회: MLPRAE '24: Proceedings of the International Conference on Machine Learning, Pattern Recognition and Automation Engineering

As network technology rapidly advances and internet penetration rates soar, network security issues are increasingly gaining public attention. This not only necessitates that network security situational awareness technology accurately identifies network attacks and assesses the current security status in complex network environments, but also involves forecasting future network security scenarios based on past data. Traditional network security situational awareness models often suffer from having too many parameters and complex models, resulting in poor timeliness, low accuracy, and weak robustness when handling large and feature-rich network data. To tackle these challenges, this paper introduces a situational prediction model for network security utilizing Selective Attention Temporal Learning (SATL). This model leverages a selective attention mechanism to effectively focus on key features. To manage the intricate and rapidly changing nature of network security scenarios and the problem of delayed prediction outcomes, this paper combines Temporal Convolutional Networks (TCN) with a selective attention mechanism to build a prediction model capable of dynamically adjusting attention weights. By conducting deep learning analysis on historical data, this model captures potential security threats and trend changes, enhancing prediction accuracy and timeliness. Finally, we perform comprehensive testing on a publicly available dataset to evaluate the model's performance. The results indicate that the SATL model surpasses other models in both error rate and accuracy, proving its superior effectiveness in forecasting network security scenarios.