A general strategy for hidden markov chain parameterisation in composite feature-spaces

A general technique for the construction of hidden Markov models (HMMs) from multiple-variable time-series observations in noisy experimental environments is set out. The proposed methodology provides an ICA-based feature-selection technique for determining the number, and the transition sequence, of underlying hidden states, along with the statistics of the observed-state emission characteristics. In retaining correlation information between features, the method is potentially far more general than Gaussian mixture model HMM parameterisation methods such as Baum-Welch re-estimation, to which we demonstrate our method reduces when an arbitrary separation of features, or an experimentally-limited feature-space is imposed.