Table 3. Canonical action research... | Emerald Publishing

Table 3.

Canonical action research cycles

Cycle	Diagnosis	Planning	Action	Evaluation	Reflection
1) Data Model Development (December 2022–mid-2023)20 weekly meetings; six interviews	Identification of missing data and communication gaps	Redesign of data model; adoption of diabetology dataset as reference	Iterative revision of model; weekly meetings; training sessions	Assessment of adequacy across departments; documentation via meetings/interviews	Recognition of knowledge proximity gap; decision to realign trajectory
2) AI Prototyping with Synthetic Data and FL (late 2023–2024)22 weekly meetings; seven interviews	Recognition of limited data availability and cross-specialty heterogeneity	Design of CTGAN-based synthetic data strategy and FL framework	Creation of local datasets; federated training rounds; harmonization of test set	Technical evaluation (accuracy, AUC and F1) and clinical interpretation	Collective reflection through interdisciplinary discussions; retraining and feature refinements
3) Clinical validation and model explanation (2025–ongoing)13 weekly meetings; four interviews	Identification of interpretability and validation challenges	Establishment of clinical explanation rounds with structured templates	Weekly sessions using SHAP plots; presentation of patient-level predictions	Collective review of variable contributions and clinical plausibility	Feedback loop enabling retraining, reweighting and ethical alignment

Cycle	Diagnosis	Planning	Action	Evaluation	Reflection
1) Data Model Development (December 2022–mid-2023)20 weekly meetings; six interviews	Identification of missing data and communication gaps	Redesign of data model; adoption of diabetology dataset as reference	Iterative revision of model; weekly meetings; training sessions	Assessment of adequacy across departments; documentation via meetings/interviews	Recognition of knowledge proximity gap; decision to realign trajectory
2) AI Prototyping with Synthetic Data and FL (late 2023–2024)22 weekly meetings; seven interviews	Recognition of limited data availability and cross-specialty heterogeneity	Design of CTGAN-based synthetic data strategy and FL framework	Creation of local datasets; federated training rounds; harmonization of test set	Technical evaluation (accuracy, AUC and F1) and clinical interpretation	Collective reflection through interdisciplinary discussions; retraining and feature refinements
3) Clinical validation and model explanation (2025–ongoing)13 weekly meetings; four interviews	Identification of interpretability and validation challenges	Establishment of clinical explanation rounds with structured templates	Weekly sessions using SHAP plots; presentation of patient-level predictions	Collective review of variable contributions and clinical plausibility	Feedback loop enabling retraining, reweighting and ethical alignment

Source(s): Authors’ own work