ODPrediction

Problem Definition. Given the regional urban characteristics of the city ${\lbrace} X_r | r\in\mathcal{R} \rbrace$ and observed OD flows $\lbrace f_{ij}|\langle r_i, r_j\rangle\in\mathcal{X} \rbrace$ between part of OD pairs $\mathcal{X}$ , construct a model to predict the remaining unknown OD flows $\lbrace f_{ij}|\langle r_i,r_j\rangle\notin\mathcal{X}\rbrace$.

Requirements

• python 3.8
• tqdm == 4.64.0
• pytorch == 2.0.0
• dgl == 0.8
• pandas == 1.4.4
• geopandas == 0.12.2
• matplotlib == 3.5.3
• mlflow == 2.3.2
• networkx == 2.8.6
• pyproj == 3.4.1
• scikit-learn == 1.1.2
• scipy == 1.9.1
• tensorboard == 2.12.3

Systematic Summary

Models	Techniques	Required Features	Feature Type
gravity	Physical Model	Population, distance	Numerical
IOM	Social Model	Opportunities	Numerical
radiation model	Physical Model	Population	Numerical
SVR	Kernal-based Model	Socioeconomics distance	Numerical
GBRT	Tree-based Model	Socioeconomics	Numerical categorical
Random Forest	Tree-based Model	Socioeconomics	Numerical
ANN	Neural Network	Socioeconomics	Numerical
SI-GCN	Deep Learning	Socioeconomics	Numerical categorical
GMEL	Deep Learning	Socioeconomics	Numerical
GCN-MLP	Deep Learning	POIs	Numerical
spatialGAT	Deep Learning	Population road density POIs railway users	Numerical
ConvGCN-RF	Deep Learning	Population landuse	Numerical categorical
SIRI	Deep Learning Causal Inference	Socioeconomics POIs	Numerical

Performance Comparison

Models	RMSE	MAE	CPC
gravity	6.944	2.179	0.602
random forest	6.273	2.436	0.638
GBRT	5.454	1.974	0.707
XGB	5.726	1.998	0.689
ANN	5.503	2.001	0.708
GNN	5.026	1.773	0.722
GMEL	4.887	1.747	0.741

The complete performance table is coming soon.