Table 4 Predicting the urban facility accessibility of out-of-sample neighborhoods.

From: Counterfactual mobility network embedding reveals prevalent accessibility gaps in U.S. cities

Category

Input

NY

LA

Chi

Dal

Hou

DC

Art & Recreation

mobility statistics

14.13%

3.04%

9.72%

8.17%

− 2.50%

12.74%

 

LINE embedding

19.58%

4.44%

15.70%

8.40%

1.63%

20.00%

 

node2vec embedding

22.56%

4.72%

26.36%

7.68%

2.10%

18.84%

 

CRANE embedding

37.18%

3.65%

33.66%

8.56%

2.43%

20.27%

Sports

mobility statistics

5.79%

0.57%

6.89%

5.16%

5.94%

3.35%

 

LINE embedding

4.97%

1.01%

7.61%

6.66%

7.92%

4.56%

 

node2vec embedding

4.71%

0.88%

7.12%

6.71%

8.17%

4.16%

 

CRANE embedding

5.98%

1.06%

7.94%

7.29%

8.36%

5.03%

Education

mobility statistics

17.90%

7.79%

24.23%

8.58%

4.85%

10.54%

 

LINE embedding

20.72%

10.66%

30.16%

9.20%

10.26%

7.79%

 

node2vec embedding

23.49%

11.01%

28.15%

12.28%

11.25%

13.29%

 

CRANE embedding

35.39%

11.95%

31.54%

14.99%

10.16%

14.27%

Health

mobility statistics

95.45%

277.39%

18.15%

27.07%

25.51%

39.90%

 

LINE embedding

78.44%

292.94%

21.61%

15.83%

34.46%

33.15%

 

node2vec embedding

100.61%

266.29%

19.39%

22.07%

36.78%

33.70%

 

CRANE embedding

105.14%

276.85%

22.36%

33.84%

44.34%

32.09%

  1. The values show the relative improvement in explained variance compared to using raw demographic features as model input. Bold text indicates the most improved method for prediction performance.