Abstract: This study employs machine learning methods to directly predict the compression-flexure capacity, tension-flexure capacity, and deformation capacity of the plastic hinge regions of double-skin steel plate concrete composite shear wall (DSCW) members under seismic actions based on their basic design parameters. First, the traditional plastic stress distribution method and machine learning methods were compared in terms of their differences in predicting the mechanical properties of structural members. Then, a fiber-based nonlinear finite element model combined with Latin hypercube sampling was used to generate a DSCW aseismic performance dataset covering a wide range of design parameters and axial load levels. Subsequently, two machine learning models were developed to predict the characteristic points of the moment–rotation hysteretic skeleton curves of the DSCW plastic hinge regions, and of the axial load–flexural capacity and of the axial load–rotation relationship curves, respectively. To reveal the effects and interactions of different design parameters on the aseismic performance of DSCWs, the Shapley additive explanations method was introduced for model interpretations. The analysis results show that the axial load ratio is the key factor governing the aseismic performance of DSCWs; however, its effect is not independent but interacts significantly with other design parameters. The findings can provide a reference for the rapid aseismic performance evaluation and for the parameter optimization design of flexure-dominated double-skin steel plate concrete composite shear wall members.