A network security situation prediction approach based on MAML and BiGRU

Network security situation prediction is a complex task that typically requires extensive retraining of deep-learning models on vast amounts of sample data to achieve optimal performance. This paper proposes an innovative approach that integrates Model-Agnostic Meta-Learning (MAML) with Bidirectional Gated Recurrent Units (BiGRU) to address these challenges. Our method harnesses the BiGRU model’s capability to learn from both preceding and succeeding conditions within network security prediction data, effectively extracting temporal information essential for prediction. This is complemented by Stochastic Gradient Descent for parameter updates, enhancing the model’s adaptability and learning efficiency. Furthermore, the MAML algorithm is incorporated to facilitate the BiGRU model’s swift adaptation to new tasks, thereby improving its generalization capabilities. The parameters are refined through a meta-learning process that calculates the sum of losses across multiple training instances and employs quadratic gradient descent for optimization. The empirical results of our approach demonstrate significant advancements, with goodness-of-fit decision coefficients of 0.926983 and 0.934452, representing a marked improvement of at least 18.0% and 15.8% over conventional deep learning models in the domain of network security situation prediction. This research novelty lies in the synergistic combination of MAML and BiGRU, which not only reduces the dependency on large datasets for retraining but also enhances the model’s predictive accuracy and generalization to novel network security scenarios. It contributes a robust and efficient solution to the critical problem of network security situation prediction and paves the way for future advancements in cybersecurity defense mechanisms.