Figure 2. Hidden Markov model labeling of maternal behavior.

An HMM is characterized by a state transition probability matrix A, an observation probability matrix B, and a state probability matrix π. The Baum-Welch algorithm is used to calculate a final HMM from an initial HMM (with user-defined, estimated probabilities) by maximizing the likelihood of emitting the observed behavioral sequence data. The final HMM is then used to label the behavioral sequence of each subject by application of the Viterbi algorithm. A statistical assessment of the frequency, duration, composition, and transitions between these labels can then be used to document strain differences in behavior.