Emerging wurtzite ferroelectric materials have aroused significant interest due to their high spontaneous polarization magnitude (P_s). However, there is a limited understanding of the key factors that influence P_s. Herein, a machine-learning regression model is developed to predict the P_s using a dataset consisting of 40 binary and 89 simple ternary wurtzite materials. Features are extracted based on elemental properties, crystal parameters and electronic properties. Feature selection is carried out using the Boruta algorithm and distance correlation analysis, resulting in a comprehensive machine learning model. Furthermore, SHapley Additive exPlanations analysis identifies the average cation-ion potential (IPi_Aave) and the lattice parameter (a) as significant determinants of P_s, with IPi_Aave having the most prominent effect. A lower IPi_Aave corresponds to a lower P_s in the material. Additionally, a exhibits an approximately negative correlation with P_s.

This multifactorial analysis fills the existing gap in understanding the determinants of P_s, and makes a foundational contribution to the evaluating emerging wurtzite materials and expediting the discovery of high-performance ferroelectric materials.

The dataset in this work can be accessed in the Scientific Data Bank https://www.doi.org/10.57760/sciencedb.j00213.00073.