Mining sequential activity–travel patterns for individual-level human activity prediction using Bayesian networks

In the past decade or so, advances in positioning technologies for capturing individual movement have given rise to a wide range of studies, including transportation, tourism, and public health. In particular, considerable effort has been made to characterize human activity–travel patterns from space–time trajectories. In contrast to visualization, geometric, or statistical methods, we propose an approach based on sequential pattern mining for analyzing human activity–travel patterns. To quantify the differences between individuals and population subgroups, we first develop a single sequential similarity measure for assessing the differences between two activity–travel patterns, then extend it to the group level with the capability to compute on two pattern sets. We also develop and implement three Bayesian network models with specially designed topology to predict the forthcoming activity at the individual level. The proposed method achieves similar or better prediction accuracy and is more robust for exploring imbalanced or sparse datasets when compared to other machine learning algorithms.