Background: We aimed to develop and validate a multivariable model for predicting dropout from a community-based going-out program for older adults. Methods: The National Center for Geriatric and Gerontology’s Study of Geriatric Syndromes, with a prospective cohort of general older adults, was employed. A total of 5905 older adults who were independent in their activities of daily living were recruited and randomly allocated into training, validation, and testing data sets in a 6:2:2 ratio. An outcome was defined as dropping out of a community-based going-out program within 180 days. An extreme gradient boosting algorithm was used to develop the predictive model using training and validation data sets and to identify feature importance. The model’s discrimination and calibration were evaluated in the test data set. Results: The area under the curve for the receiver operating characteristic ( 95% CI) was 0.701 (0.670–0.732). Sensitivity, specificity, positive predictive value, and negative predictive value and their 95% CIs were 0.253 (0.226–0.277), 0.915 (0.899–0.931), 0.588 (0.560–0.612), and 0.718 (0.692–0.743), respectively, and the slope of the calibration plot was 1.046 (0.909–1.182). Cognitive and physical functions and willingness to engage in exercise/sport activities were selected as the most important features. Conclusions: The predictive model reliably indicates whether a participant will drop out when classified as negative but not when classified as positive. Physical and cognitive functions and willingness to engage in physical activity may be primary predictors.