Explainable AI for Automated Operations: Transparent Model Insights in Dynamic Infrastructure Management

Modern infrastructure management increasingly relies on machine-learning models for autoscaling, anomaly detection, and root-cause analysis, in dynamic multi-cloud and on-premises environments. However, these models are often “black boxes” leaving operators without insight into their decisions and undermining trust and troubleshooting. I would like to propose XAI-Ops, a framework that integrates post-hoc explanation methods (SHAP, LIME), surrogate rule extraction, and interactive dashboards into automated pipelines. In a case study on Kubernetes cluster autoscaling and fault-prediction models, XAI-Ops achieved: 87 average explanation fidelity (vs. surrogate baseline 0.75) 30 % reduction in mean time to resolution (MTTR) during simulated incidents 25 % increase in operator trust scores (survey-based) < 7 ms overhead per inference for explanation generation This paper presents framework architecture, explanation techniques, integration with CI/CD, quantitative evaluation, and deployment considerations.