They found that the algorithm with the best correlation does not always have the best selection rate. Instead of building a regression curve that fits all the training data, classification ...