Table 3 Relative importance GSV features, Can-ALE metrics for log-transformed walk-to-work rates, all cities combined.

From: Predicting walking-to-work using street-level imagery and deep learning in seven Canadian cities

 

Relative importance, adjusted R2 (95% CI)

Percent of variation in walk-to-work rates explained by GSV features

City

2.8 (2.7, 3.0)

Person OD + Person OD2

14.4 (14.2, 14.6)

Building IS + Building IS2

11.9 (11.7, 12.1)

Sky IS + Sky IS2

19.7 (19.5, 19.9)

All factors combined

48.8 (48.1, 49.6)

Percent of variation in walk-to-work rates explained by Can-ALE metrics

City

3.6 (3.4, 3.7)

Street intersections

6.3 (6.2, 6.4)

Transit stops

9.3 (9.1, 9.5)

Dwellings

9.5 (9.3, 9.6)

Points of interest

11.2 (11.0, 11.3)

All factors combined

39.8 (39.1, 40.4)

  1. The R2 was calculated from linear regression models that included the variables indicated. The 95% confidence intervals of the R2 increments were estimated by sampling with replacement using 1000 bootstrap replicates. OD object detection, IS image segmentation, and CI confidence interval.
Back to article page