Rubber is widely used in applications such as tires and seismic isolation devices due to its excellent vibration damping properties. In the design of structures utilizing these materials, analyses based on hyperelastic and viscoelastic constitutive models are commonly employed. In conventional approaches, it is necessary to select an appropriate material model based on experimental data in order to represent the material behavior. In hyperelastic constitutive modeling, increasing the number of terms is a commonly adopted approach to improve accuracy. However, even when extending a two-term model to a five-term model, the material behavior observed in experiments cannot always be reproduced with sufficient accuracy. This limitation is particularly pronounced in the low- and high-strain regimes. Furthermore, while higher-order models potentially offer improved accuracy, they often suffer from reduced interpretability. Consequently, in practical applications, two-term models that can qualitatively capture the material behavior are frequently adopted. As a result, there is a strong demand for constitutive models that achieve both higher accuracy and interpretability. Here, interpretability refers to the physical meaning of model parameters. In recent years, attempts to develop a machine-learning-based constitutive model have been actively pursued [1-3]. Since machine learning can accurately predict highly nonlinear relationships, it has the potential to capture low- and high-strain behavior that is difficult to describe using conventional constitutive models. This study aims to develop a machine-learning-based constitutive model with high accuracy while considering interpretability. Numerical results obtained from analyses based on the Mooney model using the simulation software Marc are used to construct a constitutive model for rubber materials [4]. As a result, the proposed model is confirmed to reproduce stress responses with high accuracy. The effectiveness of the proposed approach is discussed.