AWS IoT SiteWise now supports retraining of anomaly detection models

Today, AWS announced new capabilities for native anomaly detection in AWS IoT SiteWise. This release includes automated model retraining, flexible promotion modes, and exposed model metrics, all designed to enhance the anomaly detection feature. The automated retraining capability allows models to be automatically retrained on a schedule ranging from a minimum of 30 days to a maximum of one year, eliminating the need to manually retrain models. This feature ensures that models stay up-to-date with changing equipment conditions or configurations, thereby maintaining optimal performance over time. Additionally, flexible promotion modes give customers the choice between service-managed and customer-managed model promotion. Automatic promotion enables AWS IoT SiteWise to evaluate and promote the best-performing model without customer intervention, while manual promotion allows customers to review comprehensive, exposed model metrics—including precision, recall, and Area Under the ROC Curve (AUC)—before deciding which model version to activate. This flexibility allows choice between a hands-off or human oversight approach. Multivariate anomaly detection is available in US East (N. Virginia) , Europe (Ireland) , and Asia Pacific (Sydney) AWS Regions where AWS IoT SiteWise is offered. To learn more, read the [launch blog](https://aws.amazon.com/blogs/industries/simplifying-industrial-equipment-monitoring-with-native-anomaly-detection-in-aws-iot-sitewise/) and [user guide](https://docs.aws.amazon.com/iot-sitewise/latest/userguide/sitewise-anomaly-detection.html).