Is My Data Safe? Predicting Instance-Level Membership Inference Success for White-box and Black-box Attacks