利用机器学习揭示高熵合金力学性能预测的定量因果效应，旨在解决成分、工艺优化中单一参数独立贡献难以被关联模型剥离量化的难题，其核心挑战不仅在于关键因素的准确识别，更在于量化该效应随其他因素变化的异质性。 本研究提出一种结合物理特征工程、机器学习建模、可解释相关性分析与因果推断的因果效应定量分析框架，实现了对高熵合金力学性能的准确预测及其异质因果效应的定量表征。 通过构造物理描述符，显著提升了高熵合金力学性能的预测精度，并识别出影响预测的关键因素为轧制变形量(CR)和退火温度(AT)，最终获得其与力学性能之间的定量因果效应。 CR 每增加 10%，抗拉强度变化范围为 -13.0~92.0 MPa(平均变化 50.1 MPa)，伸长率变化范围为 -5.02%~0.62%(平均变化 -0.29%)；AT 每升高 100 K，抗拉强度变化范围为 -164.9~-57.2 MPa(平均变化 -1 10.6 MPa)，伸长率变化范围为 5.2%~11.2%(平均变化 8.3%)。 宽的效应范围是由除关键因素外的其他弱相关因素组合的异质性决定，该效应的可靠性通过其他实验观测值验证。

To predict the mechanical properties of high-entropy alloys by machine learning, this study aimed to solve the quantitative challenge that the independent contribution of a single parameter to process control is difficult to isolate from correlated models. The core challenge of identifying quantitative causal effects lies not only in accurately identifying key factors but also in quantifying the heterogeneity of these effects as they vary with respect to other factors. A quantitative causal effect analysis framework that integrates physics-based feature engineering, machine learning modelling, interpretable correlation analysis, and causal inference was proposed in this study, enabling accurate prediction of the mechanical properties of high-entropy alloys and quantitative characterization of heterogeneous causal effects. By constructing physical descriptors, the prediction accuracy is significantly enhanced, and the key factors influencing the predictions are identified as the cold rolling reduction (CR) and the annealing temperature (AT). Ultimately, the quantitative causal effects of the two factors on the mechanical properties are obtained: for a 10% increase in the CR, the ultimate tensile strength varies between -13.0 and 92.0 MPa (with an average change of 50.1 MPa), and the elongation varies between -5.02% and 0.62% (with an average change of -0.29%); for a 100 K increase in AT, the ultimate tensile strength varies between -164.9 and -57.2 MPa (with an average change of -1 10.6 MPa), and the elongation varies between 5.2% and 11.2% (with an average change of 8.3% ). The wide range of effects is determined by the heterogeneity arising from the combinations of other weakly correlated factors beyond the key factors, and the reliability of these effects is validated through other experimental observations.