Final Results
Several variations from three families of classifiers (discriminant, probability density estimation, and
MM) were tested. Each method had some characteristics suited to the problem, but also had
flaws related to the problem.
| Classification Method | Rate | Best Features |
| Bayes | 52.20% | 8, 14, 17, 19 |
| GLD | 56.60% | 5, 6, 10, 12, 17 |
| Fischer Linear Disc. | 61.03% | 2, 8, 12, 16, 18 |
| kNN | 59.60% | 4, 15, 16 |
| Parzen Windows | 59.60% | 1, 3, 11 |
| 1st Order Markov Model | 52.92% | n/a |
| 1st Order MM (remap) | 50.78% | n/a |
| 2nd Order MM | 56.17% | n/a |
| 2nd Order MM (remap) | 55.80% | n/a |
| Hidden Markov Model | 58.60% | n/a |
Inspiration
Das EFX, Godfried Toussaint, The Neistat Brothers, Kameoka and Kuriyagawa,
Kevin Mashinter, everyone at the Media Lab 2004
Index