Feature subset selection in data-stream environments using asymmetric hidden Markov models and novelty detection