| Popis: |
In this paper we focus on a machine learning framework used in the exit analysis of crowd flow developed in Grim et al. (Analysis of Simulated Crowd Flow Exit Data: Visualization, Panic Detection and Exit Time Convergence, Attribution and Estimation. In: Gasparovic E, Domeniconi (Editors), Research in Data Science, Associations for Women in Mathematics Series, Springer, 2019, pp. 239–281). We describe a methodology for feature generation called the “sliding window technique.” In particular, we compare two unsupervised dimensionality reduction techniques such as auto-encoders and principal component analysis (PCA) and combined with a multi-output feed forward neural networks for estimating the exit times of 100 agents in a given room configuration. |
