RADAR: Robust anomaly detection and recognition. A unified and reproducible framework across tabular, time-series, and federated data

Anomaly detection is an extensively investigated research area; however, empirical progress in the field remains limited by a fragmented ecosystem of libraries, heterogeneous APIs, and inconsistent experimental workflows, which complicate fair comparison between methods and hinder experimental reproducibility. We present RADAR (Robust Anomaly Detection and Recognition), an open-source Python framework that provides a unified environment for the reproducible development, execution, benchmarking and evaluation of anomaly detection experiments across tabular, time-series and federated data. RADAR integrates state-of-the-art algorithms through a common abstraction based on a lightweight base class, enabling built-in, third-party and user-defined models to operate within a single coherent pipeline that ensures consistent experimental conditions and fair, systematic comparisons. The framework includes curated public datasets, preprocessing utilities, standardized evaluation metrics and visualization tools designed to support transparent and repeatable benchmarking, together with a graphical interface for rapid, code-free prototyping that offers the same functionality as the programmatic API. Distributed via PyPI under an AGPL-3.0 license, RADAR is designed to support transparent, reproducible and extensible experimentation, providing a practical foundation for machine learning researchers and practitioners seeking to implement, compare and benchmark anomaly detection methods in a systematic and consistent manner. • Unified anomaly-detection framework supporting tabular, time-series and federated data. • Broad method coverage by combining native implementations with carefully selected third-party detectors. • Reproducible, reusable pipelines that make experiments easier to replicate, extend and compare. • Node-based graphical workflow for rapid prototyping and low-friction benchmarking. • End-to-end environment including preprocessing, metrics, and visualization modules per data type. • Extensible design that lowers the barrier to integrating and validating new anomaly detectors.