I'm no expert on this, but my first thought would be to run a principle component analysis on the predictor variables, then use the resulting principle components to predict your dependent variable.