Simply stated, being right more than one is wrong is the same thing as being more accurate than a random assignment (as we approach a large number of "trials") if the outcomes are equally likely. And that is the point - race tells you nothing more significant about class then guessing does (if we remove 'middle class' as a selection), and class tells you nothing significant about behavior.
Also, you seem to be conflating 'class' and 'culture' now. Culture is a much, much better indicator of behavior than class, as culture speaks to beliefs and value judgements. As such, they are not interchangable ideas (that is, the propostion that 'Skin color is a good predictor of culture' is not the same as the one you made prior, 'Skin color is a good predictor of class' - but is probably just as fallacious).
Clothing and hair (God only knows what you meant by 'accent.' Care to clarify?) in isolation are strong enough indicators of culture, that race quite literally adds nothing to their accuracy - the same can be said of 'style of dress' and 'price of car driven,' for class. Race is superfluous.