We fit the well-known Hidden Markov Model (HMM) to patient arrivals data, inputted as a discrete data trace, collected over many months. The processing of the data trace makes uses a simple binning technique, followed by clustering, before it is inputted into the Baum-Welch algorithm. Upon convergence, the HMM parameters are used to predict its own synthetic traces of patient arrivals, therefore behaving as a fluid input model. Utilizing the Viterbi algorithm, one can decode the meaning of the hidden states of the HMM, further understanding the varying rate of patient arrivals at different times of the hospital schedule. Finally, an efficient set up is explored to provide optimal parameter initialization for the HMM, including choosing the number of hidden states. We conclude with a summary of our findings, comparing results with other work in the field, and extending our research in future work.
pubs.doc.ic.ac.uk: built & maintained by Ashok Argent-Katwala.