flowengineR: A Modular and Extensible Framework for Fair and Reproducible Workflow Design in R

TL;DR

flowengineR presents an R-native framework for modular, reproducible, and fairness-aware machine learning workflows by introducing a control object and interchangeable engines that cover data splitting, preprocessing, inprocessing, postprocessing, evaluation, and reporting. The design emphasizes class-free interfaces, standardized I/O, and auditable results, enabling plug-and-play fairness interventions and easy extension via templates and LLM-assisted engine development. A concrete use case on a synthetic banking dataset demonstrates fair and transparent evaluation using metrics such as $SPD$ and $MSE$, with negligible overhead attributed to the modular architecture. The framework aims to generalize beyond fairness to explainability, robustness, and regulatory compliance, and to integrate with existing R ecosystems like mlr3 and targets through a unifying, extensible workflow infrastructure.

Abstract

flowengineR is an R package designed to provide a modular and extensible framework for building reproducible algorithmic workflows for general-purpose machine learning pipelines. It is motivated by the rapidly evolving field of algorithmic fairness where new metrics, mitigation strategies, and machine learning methods continuously emerge. A central challenge in fairness, but also far beyond, is that existing toolkits either focus narrowly on single interventions or treat reproducibility and extensibility as secondary considerations rather than core design principles. flowengineR addresses this by introducing a unified architecture of standardized engines for data splitting, execution, preprocessing, training, inprocessing, postprocessing, evaluation, and reporting. Each engine encapsulates one methodological task yet communicates via a lightweight interface, ensuring workflows remain transparent, auditable, and easily extensible. Although implemented in R, flowengineR builds on ideas from workflow languages (CWL, YAWL), graph-oriented visual programming languages (KNIME), and R frameworks (BatchJobs, batchtools). Its emphasis, however, is less on orchestrating engines for resilient parallel execution but rather on the straightforward setup and management of distinct engines as data structures. This orthogonalization enables distributed responsibilities, independent development, and streamlined integration. In fairness context, by structuring fairness methods as interchangeable engines, flowengineR lets researchers integrate, compare, and evaluate interventions across the modeling pipeline. At the same time, the architecture generalizes to explainability, robustness, and compliance metrics without core modifications. While motivated by fairness, it ultimately provides a general infrastructure for any workflow context where reproducibility, transparency, and extensibility are essential.

Figures (6)

• Figure 1: Key desirable properties for sustainable workflow frameworks. These desirables summarize the research gap and form the architectural foundation of flowengineR.
• Figure 2: Comparative mapping of existing workflow and fairness frameworks against the seven desirables ( -- ). Legend: Green = property fulfilled, Yellow = partially fulfilled, Red = not fulfilled
• Figure 3: Mapping of design principles (D1--D12) to desirables (S1--S7). Circles indicate relationships, with shading encoding strength: primary (filled), supporting (unfilled), and none.
• Figure 4: Modular Workflow Structure of flowengineR. The figure illustrates the core engine types (Splitter, Preprocessing, Inprocessing, Training, Postprocessing, Evaluation, Reporting, Publishing) and their interaction within a workflow pipeline. This modular design enables clear separation of concerns and standardized interfaces across all components.
• Figure 5: Extensibility through standardized engine interfaces. The figure shows how new engines can be added independently without modifying the core system. This openness allows community-driven contributions and long-term scalability of the framework.