A Method for Network Security Situation Assessment Based on Multi-source Feature Fusion in Big Data

Network security status assessment is a critical research area. Traditional methods like Dempster-Shafer evidence theory and Bayesian algorithms have limitations such as device specificity and low data dimensionality. Machine learning algorithms, while promising, are often limited by poor generalizability and data source dependence. This paper introduces a novel multi-source feature fusion evaluation system based on denoising deep autoencoder (DAE) networks. The system employs a three-step approach: (1) feature extraction from diverse data sources using denoising DAE, (2) linear feature fusion via concatenation, and (3) network security situation analysis and prediction with a random forest model. The overall network security situation is calculated by integrating attack probability and impact. Our system outperforms three existing methods, including the backpropagation neural network, with an accuracy rate of 94.96%. This suggests that our multi-source feature fusion approach provides a more robust and accurate method for network security status evaluation.