It handles more variables (up to 50,000 in many cases), does not inject noise into the input data by requiring you to fill in missing data with assumed values, and it allows for small patterns to be recognized and not smoothed out.