Fig. 5: Model performance at predicting flows at small distances.
From: Human mobility is well described by closed-form gravity-like models learned automatically from data
We consider the data used by Simini et al.12 about mobility flows between census tracts within small geographical (25 × 25 km2) regions in NY state. We evaluate predictions over test data using the same metrics as in Fig. 3: A Common part of commuters; B Absolute error; C Absolute relative error; D Absolute log-ratio. In each panel, triangles (◂) indicate the best-performing model according to the corresponding metric; BMS models are the best-performing in all metrics. See Methods and text, for a complete description of the metrics, the models, and the data.
