AI-Driven Performance Evaluation of Network Digital Twins for 6G: A Unified Taxonomy and Polymetric Twinning Index

This paper addresses the challenge of evaluating Network Digital Twins (NDTs) and discusses how to evaluate the performance of NDTs in AI-driven 6G systems. The paper makes three main contributions. First, it provides a layer-by-layer analysis of enabling technologies supporting NDTs, examining how tools like edge computing, blockchain, federated learning, and various data collection methods fulfill specific NDT requirements across key characteristics of NDTs. Building on this, it proposes a comprehensive taxonomy of performance evaluation metrics. This taxonomy systematically classifies both digital twin-specific metrics and general performance metrics. Finally, it introduces the polymetric twinning index, a simple, unified metric that aggregates multiple performance indicators into a single score. The approach is validated through five case studies spanning IoT and 5G networks, demonstrating how different metric subsets can be selected based on specific NDT focus areas. Results show that the taxonomy and polymetric twinning index together form an extensible framework for assessing NDTs across domains, which enables fair baseline comparisons when appropriate subsets are chosen.