HMMs have been successfully applied for many tasks related to sequence
clustering [18], such as map matching with noisy observations [15], and to hidden
intents recognition [1, 19]. We also dier from [4], where a traditional Markov
model is used to classify dierent observations to an internal taxonomy. The work
proposes a one-to-one mapping to couple state-symbol, but due to the absence of
hidden states, Baum-Welch and Viterbi algorithms are not applied. Our proposal
is the rst attempt to addresses the problem of predicting the destination and
the route to reach it with a HMM formulation over noisy observations.