You are on page 1of 1

Independent multi-class classi er

This classi er predicts each letter separately. Therefore each prediction is independent
and therefore this classi er has the weakest performance of the three classi ers used.

+ Can predict any word while learning from completely di erent training words, The
performance remains the same while using a test set with di erent words

-A lot of misclassi cations

Structured, pair-wise dependency


This classier predicts each letter with respect to its neighbors. Therefore by computing
the probabilities of neighboring characters, the performance improves

+ Better than independent classi er, works on a testing dataset that di ers from training
dataset (usually with worse results)

- If the test dataset contains di erent words than the training dataset, the performance
can get worse drastically

Structured, xed number of sequences


This classi er predicts the sequences as a whole.

+ Lowest error rate during testing

- Can not classify words that were not present in the training dataset

Testing errors in %

Rseq Rchar
independent multi-class classi er 70 27

structured, pair-wise dependency 10 5

structured, xed number of sequences 2 2


fi
fi
fi
fi
fi
fi
fi
ff
fi
fi
ff
ff
ff
fi

You might also like