Network Attack Detection Based on Causal Genetic Algorithm

To address the critical limitations of traditional machine learning-based intrusion detection systems (IDS)—including low detection rates, elevated false positives/negatives, and insufficient real-time responsiveness in emerging sophisticated attack scenarios — this study proposes an optimized intrusion detection framework integrating causal inference with genetic algorithms. Our methodology involves: (1) constructing a causal dependency graph to model the propagation mechanisms among network traffic features, quantifying causal effects on attack behaviors through counterfactual analysis; (2) implementing multi-objective feature optimization via an adaptive genetic algorithm that encodes feature subsets as binary chromosomes, with fitness functions combining detection performance metrics and causal importance weights. Experimental validation on a benchmark dataset demonstrates significant improvements. The model shows high detection accuracy while effectively reducing both false positive and negative rates. It also achieves enhanced realtime processing efficiency compared to conventional approaches. Notably, the causal inference mechanism leads to a substantial boost in accuracy for identifying zero-day attacks, validating the framework's superiority in feature selection interpretability and model generalization. This work establishes a causal-aware feature optimization paradigm for intrusion detection systems operating in dynamic adversarial environments, addressing both performance bottlenecks and the black-box nature of traditional machine learning models.