Forecasting using dynamic factor models with cluster structure at Barcelona subway stations

Abstract

Dynamic factor models are a powerful technique for analysing vast volumes of data, more precisely, time series. However, the large volumes of data that come from public transport networks tend to have heterogeneity and a cluster structure. In this paper, Dynamic Factor Models with Cluster Structure (DFMCS) are used to forecast hourly entrances in the different stations of the Barcelona subway network. The main and most novel contribution lies in the use of clustering techniques to make an initial grouping of the behaviour of the elements belonging to the time series, in order to subsequently be able to predict future patterns.