Fig. 5: Extreme event model evaluation.
From: Industrial-scale prediction of cement clinker phases using machine learning
Distribution of (a) alite, (b) belite, and (c) ferrite over two years of plant operation. Data retained after preprocessing is shown in green, while data removed during the three-tier outlier removal process is shown in red. The red-marked events correspond to rare plant operations with very low occurrence frequency, denoted as f for each phase. d–f illustrate the performance of the best-performing models across normal operation ranges (green background) and rare operation conditions (red background). The total probability of normal and rare operations is also indicated for each phase. The prediction errors for (g) normal operation and (h) extreme operation are shown using box plots of absolute errors.
