Performance comparison across hidden, pairwise and triplet Markov models’ estimators