sDNA for reference class forecasting with explainable AI

sDNA is now at the core of adapting Flyvberg‘s Reference Class Forecasting megaproject outcomes appraisal – using explainable AI to appraise new town, CBD, road and rail infrastructures:

A SEM-ANN adaptation of reference class forecasting for the assessment of large-scale urban planning vision, Land Use Policy, Zhou, Y., Zhang, L., and Chiaradia* AJF.

Available here or here.

Key items of interest are:

  1. the need to appraise early front-end visioning scenarios based on relatively sparse data
  2. infrastructure projects appraisal needs to be able to incorporate network impacts which are notoriously difficult to identify and model.

sDNA contributes analyses of existing and proposed scenarios of urban rail network (MTR), road network and pedestrian network to a structural equation model (SEM) to determine significant variables which are fed into an artificial neural net (ANN) giving a form of explainable AI. Pedestrian network is here proxied via road network. The case study is a megaproject in Hong Kong HKSAR, China.

sDNA for road travel time estimation and economic analysis

Our first foray demonstrating the use of sDNA in economics is now published:
Hossain, MA, Cooper, CHV. Spatial network analysis as a tool for measuring change in accessibility over time: Limits of transport investment as a driver for UK regional development. Growth and Change. 2021; 00: 1– 25.

One aspect of wider interest here is our proxy for road travel time, which has correlation (r2) of up to 0.99 with journey times taken from Google maps. All details in the paper so you can reapply the formula in your own network analysis.
In our case this was a national scale network analysis (3 million links) so we used selective sampling of origins to speed it up, though each origin is analysed for quantity and shape characteristics of all destinations within a 1 hour travel time radius.