Explainable AI (XAI) for Network Operations and Troubleshooting

With the advancement of 5G, Software-Defined Networking (SDN), and Network Function Virtualization (NFV) technics, which enable modern network infrastructures to become more and more complex and dynamic, the rule-based controller-based management systems are becoming insufficient to pinpoint the faults in real-time, to determine the root cause, and to conduct fine-grain optimization according to the needs of users. Artificial Intelligence (AI) especially Machine Learning (ML) has become an efficient means of automating the network. Nonetheless, the secrecy of most AI models simply acts as a serious impediment to trust, responsibility, and transparency in the operations, particularly in essential networking contexts. In the following paper, I will be examining the process of incorporating techniques of Explainable AI (XAI) into network operations and troubleshooting as a component of work. We propose a conceptual integration of interpretable machine learning models and post-hoc explanation algorithms, including SHAP and LIME into currently developed AIOps pipelines. We test the utility of XAI to identify anomalies and localise faults using real and synthetic network telemetry data and provide useful information to the network operators. Those findings demonstrate that there exists a significant trade-off between model accuracy and model interpretability but also that XAI holds the chance of considerably improving human-AI teams in daily decision-making. Our results open the path to develop auditable and transparent and resilient AI-based networks.