Utilizing XAI technique to improve autoencoder based model for computer network anomaly detection with shapley additive explanation(SHAP)

Explainability can improve model performance: The paper demonstrates that using XAI techniques (specifically SHAP) to understand which features drive anomalies led to better detection performance than traditional feature selection methods, suggesting that transparency and performance are complementary rather than competing goals.

Black-box AI poses real barriers to adoption in security: The authors identify the lack of transparency in machine learning and deep learning models as "a major obstacle to their implementation" in critical areas like network security, highlighting why explainability matters beyond academic interest—it affects whether organizations will actually deploy these systems.

SHAP can serve as a practical feature selection tool: Rather than just explaining predictions after the fact, kernel SHAP was used to identify which specific features actually cause anomalous behavior, demonstrating that XAI techniques can have direct practical applications in improving how models are built, not just understood.

Understanding model internals enables targeted improvements: By interpreting the autoencoder's outputs through SHAP, the researchers were able to retrain the model using only the most relevant features, showing that explainability can guide concrete engineering decisions that enhance trustworthiness and effectiveness.