0

I am working on handwriting digit recognition based on Hidden Markov Models(HMMs). After training, we get 5 models. I take a sample and calculate its probability likelihood with each models. Then these are the result :

  • For model 1: -235
  • For model 2: -250
  • For model 3: -193
  • For model 4: -290
  • For model 5: -325

It is shown that the sample will fall into model 3rd since it has the highest probability(-193) . The question is how can I convert the value (-193) to percentage in order to express it as confidence level of the recognition ?

SKRUY
  • 69
  • 13

1 Answers1

0

Actually those outputs are the logarithm of the likelihood that the input provided could have been generated by the corresponding HMM. That said, one measure of the confidence level of the recognition could be obtained by taking the exponencial of this log-likelihood, since it would give you a probability measure.

Diego Stéfano
  • 189
  • 2
  • 5