Identifying Classes Susceptible to Adversarial Attacks