This paper presents a machine learning model for load-level estimation for shear-critical reinforced concrete (RC) beams and slabs using multifractal features of their characteristic crack patterns to automate and provide well-informed decisions for RC damage assessment. Multifractal analysis was conducted on a database of 508 images, of which critical features were extracted from the singularity and generalized dimension spectra. These features are used as predictors for the load-level estimation model. The extreme gradient boosting algorithm yielded the best performance among the four machine learning models considered. The mean of the predicted-to-true ratio for the developed model was 1.04 with a coefficient of variation of 0.27. Upon applying Shapley additive explanations, the fractal dimension, information dimension, correlation dimension and the area under the left branch of the singularity spectrum were the critical features influencing load-level estimation. The proposed model can be useful to RC building inspectors.