This study proposes computational homogenization for elastoplastic heterogeneous beams using machine learning. To investigate the mechanical behavior of composite materials, multiscale analysis methods have been studied based on homogenization theory and some methods have been proposed for heterogeneous beams. However, most are for elastic beams, and only a few studies have considered inelastic behavior. This has been due to the complexity of the constitutive relationship between the cross-sectional deformation state and the internal forces when inelastic materials are included as constituents. Therefore, in this study, we propose a multiscale analysis method for elastoplastic heterogeneous beams by means of RBF interpolation. A dataset of cross-sectional deformation and force is generated by performing numerical experiments on a unit cell representing the periodic microstructure of a beam under various conditions, and then a continuous function of cross-sectional force is constructed by machine learning with the dataset. Multiscale analyses of elastoplastic heterogeneous beams are performed using the resulting function to compute the cross-sectional force history for an arbitrary deformation history.