Scholarlink

Rapid intrusions in modern defense networks necessitate doctrinal decision-making and detection at the millisecond level. We suggest a neuro-symbolic intrusion detection pipeline that combines rule-based reasoning through experta with ensemble machine learning (ML) (Logistic Regression, Decision Tree, Random Forest, XGBoost, SVM, k-NN). XGBoost obtained the highest accuracy (0.9998) and ROC-AUC (0.99999) using the KDD’99 dataset under 5-fold CV and Random Forest obtained the highest accuracy (0.9998) under 90:10 ratio. Three key features were identified by SHAP: src_bytes, srv_count, and same_srv_rate. MITRE ATT&CK was mapped to these (e.g., High srv_count → T1499 Endpoint DoS). The system produced readable alerts for military use and displayed a variance of less than 0.3% across runs. Real-time ingestion and quantum-kernel support for edge deployment are examples of future work.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Defending Digital Borders with Neuro-Symbolic AI: An Explainable Approach to Tactical Intrusion Detection

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sayan Pal,
  • Rahul Karmakar

摘要

Rapid intrusions in modern defense networks necessitate doctrinal decision-making and detection at the millisecond level. We suggest a neuro-symbolic intrusion detection pipeline that combines rule-based reasoning through experta with ensemble machine learning (ML) (Logistic Regression, Decision Tree, Random Forest, XGBoost, SVM, k-NN). XGBoost obtained the highest accuracy (0.9998) and ROC-AUC (0.99999) using the KDD’99 dataset under 5-fold CV and Random Forest obtained the highest accuracy (0.9998) under 90:10 ratio. Three key features were identified by SHAP: src_bytes, srv_count, and same_srv_rate. MITRE ATT&CK was mapped to these (e.g., High srv_count → T1499 Endpoint DoS). The system produced readable alerts for military use and displayed a variance of less than 0.3% across runs. Real-time ingestion and quantum-kernel support for edge deployment are examples of future work.