LSTM models for spatiotemporal extrapolation of population data

Kurzfassung

The anticipation of future geospatial population distributions is crucial for numerous application domains. Here, we capitalize upon existing gridded population time series data sets, which are provided on an open source basis globally, and implement a machine learning model tailored for time series analysis, i.e., Long Short Term Memory (LSTM) network. In detail, we harvest WorldPop population data and learn an LSTM model for anticipating population along a three-year interval. Experimental results are obtained from Peru’s capital Lima, which features a high population dynamic. To gain insights regarding the competitive performance of LSTM models in this application context, we also implement multilinear regression and Random Forest models for comparison. The results underline the usefulness of temporal models, i.e., LSTM, for forecasting gridded population data.