Modelo gravitacional versus XGBoost: aplicações para a pendularidade na Região Metropolitana de Londrina

Abstract

This research aims to analyze the determinants of commuting movements between pairs of cities in the Metropolitan Region of Londrina in 2010. To achieve this goal, a sample of 600 pairs of cities was used to model a gravitational equation with pseudo-maximum likelihood Poisson estimation and an XGBoost model. Furthermore, spatial effects were included in both models, and their predictive capacity was compared. The results indicate that distance is a repelling factor for commuting workers, while the labor market variables of the origin and destination cities are repelling and attracting factors, respectively. Spatial effects are relevant, although they do not enhance the predictive capacity of the models. Lastly, the gravitational model, after the proposed sophistication, managed to match the performance of the XGBoost given the comparison metrics and the characteristics of the studied sample.