Clustering the seoul metropolitan area by travel patterns based on a deep belief network

It is very useful to divide an urban area into several homogeneous zones when managing a city. Conventionally, clustering urban areas has depended upon the intuition and expertise of urban planners. On the other hand, the present study regards specific travel patterns of people living in an urban zone as a key variable to distinguish that zone from others. A K-means algorithm was adopted to cluster zones of the Seoul metropolitan area, after large dimensional origin-destination flows, which were elicited from smart-card data, were reduced by using a principal component analysis. A more elaborated approach, a deep belief network that stacked multiple restricted Boltzmann machines, was also used to reduce the dimension of origin-destination travel flows. The latter approach unveiled more hidden nonlinear aspects of clustering than provided by either the conventional zoning convention or the PCA-based approach.