Wang, S., Zheng, Y., Wang, G. et al. Infrequent activities predict economic outcomes in major American cities. Nat Cities

Objective:

• The fundamental link between mobility and economy is still missing
• Key: method ref

Case:

• Global cities

Methodology:

• Network
• Power law
• ML:
  • linear regression with elastic net regularization to tackle overfitting
  • Gradient boosting regressions to identify non-linearly predictive activities
• Sensitivity:
  • Gauassian noise: randomly omitting 10% of the data and dropp the data below the lower bounds and above the upper bound
  • Transferability test
  • Benchmarking frequent activities
  • Grouping activities
• Temporal regression

Data Source

• Mobility: MIT IRB office
• Socioeconomic: American community survey

Findings:

• Mobility activities can predict more than 50% of variation in income and property values

• A small fraction of infrequent activities is needed to achieve relatively high performance

Coding Reference:

• Python