I have a classifier which assigns a behavioural class to segments of data such that the output emissions that result are a sequence of predicted animal behaviours such as:
Foraging ==> Foraging ==> Resting ==> Resting ==> Relocating
Inevitably, some behaviours will be misclassified for example the likelihood of finding a Relocating segment within a resting segment would be very low within the context of my work e.g:
Resting ==> Resting ==> Relocating ==> Resting ==> Resting
Therefore I would like to work towards reducing this type of error in the classified sequences.
My question is, would it be possible to train a hidden Markov model using the transition and emission probabilities of the 3 behaviours to 'smooth' such sequence errors such that:
Resting ==> Resting ==> Resting ==> Resting ==> Resting
