A0497
Title: Spatiotemporal factor models for functional data with application to population map forecast
Authors: Tomoya Wakayama - The university of Tokyo (Japan) [presenting]
Shonosuke Sugasawa - Keio University (Japan)
Abstract: With the proliferation of mobile devices, an increasing amount of population data is being collected, and there is growing demand to use large-scale, multidimensional data in real-world situations. Functional data analysis (FDA) was introduced into the problem of predicting the hourly population of different districts of Tokyo. FDA is a methodology that treats and analyzes longitudinal data as curves, which reduces the number of parameters and makes it easier to handle high-dimensional data. Specifically, by assuming a Gaussian process, the large covariance matrix parameters of the multivariate normal distribution were avoided. In addition, the data were time and spatially-dependent between districts. To capture these characteristics, a Bayesian factor model was introduced, which modelled the time series of a small number of common factors and expressed the spatial structure in terms of factor loading matrices. Furthermore, the factor-loading matrices were made identifiable and sparse to ensure the interpretability of the model. A method for selecting factors using the Bayesian shrinkage method is also proposed. The forecast accuracy and interpretability of the proposed approach through numerical experiments and data analysis were studied. It was found that the flexibility of the proposed method could be extended to reflect further time series feature contributing ted to the accuracy.