The Optical Gravitational Lensing Experiment (OGLE) continuously monitors hundreds of thousands of eclipsing binaries in the Galactic bulge field and the Magellanic Clouds. These objects have been classified into major morphological subclasses, such as contact, noncontact, ellipsoidal, and cataclysmic variables, both by matching the light curves with predefined templates and by visual inspections. Here we present the result of a machine-learned automatic classification based on the morphology of light curves inspired by the classification of eclipsing binaries observed by the original Kepler mission. We similarly use a dimensionality reduction technique with locally linear embedding to map the high dimension of the data set into a low-dimensional embedding parameter space, while keeping the local geometry and the similarities of the neighboring data points. After three consecutive steps, we assign one parameter to each binary star, which scales well with the "detachness," i.e., the sum of the relative radii of the components. This value is in good agreement with the morphology types listed in the OGLE catalog and, along with the orbital periods, can be used to filter any morphological subtypes based on the similarity of light curves. Our open-source pipeline can be applied in a fully automatic way to any other large data set to classify binary stars.