Leveraging Inequality‐Constrained Data for Enhanced Liquidus Temperature Prediction in Nuclear Waste Glass Melts

Inequality‐constrained data are frequently discarded in engineering, leading to significant information loss in data‐scarce domains such as nuclear waste glass property modeling. This paper presents a nonparametric censored‐data regression framework based on an ‐norm optimization criterion that leverages slack variables to integrate left‐, right‐, and non‐constrained observations into training. Validation was performed on synthetic data and a neural network (NN) was developed for predicting spinel liquidus temperature () in high‐level nuclear waste glasses. This method improved the coefficient of determination () from 0.5780 to 0.8622 and reduced Mean absolute error (MAE) by 41.73% (i.e., from 54.15 to 31.56°) on deterministic values. The baseline NN failed to satisfy any inequality constraints while the proposed ‐norm NN satisfies 63.64% of the constraints. Ensemble analysis further showed improved uncertainty calibration, reducing Mean Absolute Calibration Error (MACE) from 0.2018 to 0.0906. The proposed framework effectively extracts actionable information from previously unusable data to enhance predictive accuracy, reduce epistemic uncertainty, and improve physical consistency in complex industrial applications.