EXPLAIN-IT: Towards Explainable AI for Unsupervised Network Traffic Analysis

Clustering algorithms alone don't tell the full story: The paper demonstrates that unsupervised learning techniques, while useful for discovering patterns in network data, often fail to provide clear explanations for their results—a critical gap when these systems influence real-world decisions about network management and security.

Explainability can be extended beyond supervised to unsupervised learning: EXPLAIN-IT introduces a methodology that bridges a technical gap by applying explainable AI principles to clustering tasks, showing that transparency isn't limited to traditional supervised models and can help users understand why data was grouped in particular ways.

Real-world application: encrypted traffic analysis: The paper demonstrates practical relevance by applying their approach to YouTube video quality classification in encrypted network traffic, suggesting that explainable clustering has concrete applications in network monitoring where understanding why classifications occur matters for trust and debugging.

Transparency enables better human oversight: By providing explanations for clustering decisions on complex, unlabeled data, EXPLAIN-IT supports more responsible AI deployment—allowing network administrators and security teams to verify that automated groupings make sense rather than blindly trusting algorithmic outputs.

Complexity requires explanation: The paper emphasizes that when input data has complex structures and distributions (common in real network traffic), explainability becomes essential for practitioners to validate whether the AI system is capturing meaningful patterns or artifacts.