Different Horses for Different Courses: Comparing Bias Mitigation Algorithms in ML