We study the tree-based interpretable machine learning of the thermodynamic phases in a square lattice model. By this method, the interpretability can be achieved and the precision is very high. We obtain the influence of each input feature and catch the main contribution to thermal equilibrium states, without the prior knowledge of phase classification and transition. The tree-based interpretable machine learning can be used to study the unclear impact of inputs on the physical properties and distinguish the roles of input features playing.

中文翻译：

---

热力学阶段的基于树的可解释机器学习

我们研究了方形晶格模型中热力学阶段的基于树的可解释机器学习。通过这种方法，可以实现可解释性，并且精度非常高。我们获得每个输入特征的影响并捕捉对热平衡状态的主要贡献，而无需相位分类和转变的先验知识。基于树的可解释机器学习可用于研究输入对物理特性的不明确影响，并区分输入特征发挥的作用。