Abstract
Enterprise information systems contain interconnected applications, APIs, databases, cloud resources, and infrastructure whose performance can change under varying workloads and service conditions. Conventional monitoring provides metrics, logs, traces, and events, but these signals often offer limited support for future performance prediction, dependency analysis, and operational response. This study presents a Digital Twin and Software Observability Framework for predictive performance management in enterprise information systems. The framework combines multimodal telemetry with a dynamic Digital Twin representation of service health, workload conditions, performance indicators, configuration information, and dependencies. Machine-learning models estimate abnormal behavior and service degradation risk, while a dependency-aware decision module evaluates potential downstream effects and recommends operational actions. Evaluation uses Train Ticket, Online Boutique, and Sock Shop microservice environments, public failure datasets, synthetic workloads, and controlled fault scenarios. The proposed framework achieved a Precision of 0.914, Recall of 0.902, F1-score of 0.908, and AC@k of 0.861, with an MTTD of 11.4 seconds. Results indicate that the integrated framework supports predictive diagnosis and operational decision making under incomplete trace coverage.