Figure 9
The H S C M L model quality dimensions comprise accuracy, computational efficiency and responsiveness, robustness and generalisability, solution optimality, explainability and interpretability, and uncertainty quantifiability. Accuracy measures the gap between predicted and actual values using numerical error, classification, spatial accuracy, structural similarity, geometric accuracy and ranking correlation metrics. Computational efficiency and responsiveness consider inference, initialisation, training and tuning time. Robustness and generalisability consider training monitoring, overfitting control and performance across imperfect or shifted data, synthetic and real-world data, and different datasets. Solution optimality assesses operational effectiveness, demand fulfilment, deadlines, latency, multi-criteria optimisation outcomes and humanitarian impact indices. Explainability and interpretability consider feature importance and traceability of calculated variables and model inputs. Uncertainty quantifiability includes confidence intervals, ensemble prediction variance, probabilistic prediction intervals and bootstrap variance estimation.Aggregated framework for HSCML model quality dimensions
Source: Authors’ own work