A preference disaggregation method compatible with scarce data for risk management of mine-derived water pollution
针对矿山源水污染历史数据稀缺的问题，提出一种凸二次偏好分解模型，将点监督转化为关系监督，实现高污染样本的预测识别，帮助优先分配监测资源。
Balancing production and environmental protection is a core challenge in mining activities, where heavy metal discharge may cause water pollution. This calls for effective environmental risk management based on pollution assessments that identify high-risk cases and provide actionable insights. Given the high cost of water quality tests, the predictive identification of high-risk samples based on historical data can streamline pollution assessments and enable the prioritisation of monitoring resources. However, the sporadic nature of mining-related pollution events may lead to scarce historical data, making predictive modelling difficult. To address data scarcity, this study develops a convex quadratic preference disaggregation model that reformulates pointwise supervision into relational supervision using the additive fuzzy preference structure. The model relaxes parametric assumptions incompatible with pollution assessment through piecewise linear approximations of nonlinear risk preferences, and supports both relative and absolute predictive outputs to suit diverse decision scenarios. A case study of mine-derived water pollution assessment with scarce data and experiments on standard datasets demonstrate the effectiveness and applicability of the method. Environmental management insights regarding sample screening, risk stratification, and deferred decision-making are provided.