Enhanced population estimation beyond counts: modelling age patterns

Abstract

Fine-scale population distribution and demographic knowledge are essential to support efficient urban monitoring, planning and decision making. In particular, urban areas face many future challenges concerning rapidly growing and ageing societies. Recent studies have extensively disaggregated population counts, while information on demographic measures including age have remained unattended. In this work, we develop methods to enhance demographic prediction by estimating not only population numbers, but also average age and the proportion of the elderly (65 years and older). By implementing and comparing different ML-algorithms (Random Forest, Support Vector Machines and Linear Regression), we show that enhanced demographic patterns can be estimated using Points-of-Interest (POI) and real estate data. Linear Regression has proven to perform best and most constantly throughout the predictions, while Random Forest and Support Vector Machines are more sensitive to the training data. Our results reveal that there is a strong relationship between a) population counts and the number of POI and information on real estate (block and transaction counts), and b) age distributions and real estate metadata (property age and flat type). This study highlights the potential of a small set of detailed, strong predictors in combination with appropriate models for population estimations beyond counts in urban areas.