AutoDrift: A Forecast-Aware Concept Drift Detection and Retraining Pipeline in MLOps with CMAPSS

Concept drift poses a critical challenge to deploying reliable machine learning models in real-world production environments, particularly in time series forecasting and predictive maintenance systems. We present AutoDrift, a modular, forecast-aware framework to detect concept drift and trigger the automated retraining of models within modern MLOps pipelines. This approach integrates Facebook Prophet for time series prediction, statistical drift detection techniques such as the Kolmogorov-Smirnov test, and Apache Airflow to orchestrate retraining and deployment tasks in response to detected drift. Using the widely bench marked NASA CMAPSS dataset, we simulate real-world engine degradation and sensor drift conditions to evaluate AutoDrift's effectiveness in predictive maintenance scenarios. The results demonstrate that AutoDrift maintains high forecasting accuracy (91 % precision) while reducing retraining latency by 37 % compared to static retraining schedules. The framework supports model versioning via MLflow and is cloud-agnostic, enabling seamless deployment across Kubernetes, AWS, and on-premises environments. AutoDrift offers a practical solution to one of the most pressing MLOps challenges such as maintaining the adaptability of deployed models to changing data distributions, by unifying drift detection, retraining logic, and orchestration in a single, extensible pipeline. The system is designed to be interpretable, scalable, and easy to integrate into existing machine learning operations, making it suitable for industrial applications in aviation, IoT, and large-scale telemetry monitoring.