Compute the proportion of incorrect predictions within a set of instances. Assumes that values in knownValues and predictedValues are categorical.