The proliferation of digital health technologies provides a novel opportunity to capture detailed human activity patterns, including sequences of time-stamped location and activity states (e.g., Home, Work, Eat Out, Car, Walk). This rich data, typically collected via smartphone app, offers significant potential for enhancing health research by serving as sensitive outcomes for interventions, vital context for physiological sensor data, and earlier predictors of adverse health conditions.

However, the sequence-based structure of these data—characterized by discrete activity states, continuous durations, high heterogeneity, and complex temporal dependencies—presents substantial biostatistical and computational challenges. This seminar showcases recently completed and ongoing methods research for characterizing, synthesizing, and predicting human activity sequences. The statistical tools underlying these methods are diverse, including sequence alignment, adjacency matrix decomposition, paired Markov models, and a little-known variant of the Lasso which performs particularly well with human activity sequence data. Several open research questions in this area will be highlighted.