Utilization of hidden Markov models for sports data analysis

This thesis applies hidden Markov models to the analysis of football match dynamics. Using event data from 35 matches of FC Viktoria Plzeň, a minute-wise categorical activity variable with three levels is constructed to capture the dominance of one team. For this time series, a two-state and a three-state hidden Markov models with state-dependent one-parameter emission distributions are proposed. A simulation study shows that the two-state model provides stable parameter estimates, while the three-state model suffers from practical non-identifiability. The empirical application to real data therefore focuses on the two-state model, which identifies a passive/neutral state and a state of strong offensive pressure.