Vulnerable atherosclerotic plaques have high risk for rupture, with rupture potential related to plaque composition and structure. We have previously shown that soft (intraplaque hemorrhage IPH, and lipid rich necrotic core LRNC) are differentiated from stiff (collagen COL, and calcium CAL) plaque elements in human carotid plaques by Acoustic Radiation Force Impulse (ARFI)-derived peak displacement (PD). However, PD had lower performance for differentiating between features with similar stiffness. Here we evaluate an alternative method to improve intraplaque feature delineation by using machine learning methods. From ARFI imaging data, SNR, cross-correlation coefficient, and displacement were used as inputs to random forests (RaF) and support vector machines (SVM) algorithms. The algorithms were trained to identify IPH, LRNC, COL and CAL by 5-fold cross-validation with ground truth identified from histology. From output likelihood matrices, CNR between plaque components were calculated and compared to the corresponding CNR achieved by ARFI PD and VoA. Results showed that both RaF and SVM achieved higher CNRs for distinguishing between features than ARFI outputs alone. These results suggest that, relative to PD, machine learning improves ARFI discrimination of carotid plaque components that are correlated to vulnerability for rupture.