Reconstructing Individual Activity Trajectories by Hidden Semi-Markov Model

The individual trajectory of human activity collected through crowdsourcing implies the characteristic regularity of individual activity. There are always data gaps, and it introduces negative influence of individual activity pattern analyses. By extracting the internal regularity of known data, the incomplete relationship between location and time can be imputed and reconstructed. In this paper, a method of trajectory reconstruction based on individual behavior pattern and Hidden Semi-Markov Model (HSMM) is presented. It introduces behavior state variables and a discrete representation of locations to facilitate explanation of real-world meaning of the model parameters in individual daily activities. In HSMM, the interaction between various parameters imitates the behavior state transition and spatiotemporal variation. The relationship between the individual locations and behavior states is captured as an individual activity pattern. This pattern is used to derive individual state change sequence in the studied period of time. The state sequence can be resolved to obtain the final individual activity trajectory. Experiment on an individual track data set including Weibo check-ins and mobile phone GPS records showed that the accuracy of the trajectory reconstruction was 84.2%.

Han, Z., Wan, Z., Guo, W., & Ren, C. (2018). Reconstructing Individual Activity Trajectories by Hidden Semi-Markov Model. In 2018 26th International Conference on Geoinformatics (pp. 1-6). IEEE.