Social grooming is a widespread behavior in many primate species that serves various social functions, including strengthening social bonds, reinforcing social hierarchies, and reducing stress. Previous studies have shown that social grooming can be identified by visual observation or manual annotation of acceleration signals collected from animal-attached sensors. However, such manual labeling is time-consuming and requires expert knowledge. To facilitate the large-scale monitoring of social grooming behavior in wild primates, we propose a machine learning approach for automatic identification of social grooming from acceleration signals. We developed a dataset of acceleration signals collected from wild baboons ( _Papio anubis_), containing over 100 hours of social grooming and over 500 hours of non-social grooming activities. The dataset was used to train and test a variety of machine learning models, including support vector machines, decision trees, and random forests. Our results show that the best model, a random forest, achieved an accuracy of 96.2% and an F1 score of 94.5% in identifying social grooming events. The proposed approach is promising for automatic and large-scale identification of social grooming behavior in wild animals, which can contribute to our understanding of primate social behavior and provide valuable information for conservation and management efforts.