Playing around with applying our method from doi.org/10.1177/239980… for a new project; we can use it to map, e.g., (2016) walking commutes by electoral division in #Dublin using flowmapper.org
We can also combine this with our cost-distance method (self-promotion! sciencedirect.com/science/articl…) to find the cost of traveling a given distance (based on opportunity cost of travel time, buying walking shoes, etc.) to see where walking infrastructure/service is relatively poor
From this it looks like more suburban/outlying areas (that likely have more disconnected street patterns) are relatively more costly to walk
Here are (again, 2016) public transit commuting flows in Dublin (left)...and transit flows with the largest 500 average waiting times at 09:00 on Nov. 29 (right)
These patterns (including overall cost-distance ratio here) have important implications for #equity, #sustainableLiving, and improving non-auto transportation across the region. It will also be interesting to see what changes the @CSOIreland 2022 Census results bring!
• • •
Missing some Tweet in this thread? You can try to
force a refresh
For all interested at the development of spatially-explicit machine learning models, new paper in @GeogAnalysis develops a straightforward approach: insert spatial lags into a random forest model and test performance against spatial econometric models onlinelibrary.wiley.com/doi/10.1111/ge…
1/8
Random forest outperforms SE models across the board, and the "spatial Durbin" RF model performs best of the lot, though not by much: 84.61% accuracy vs. 84.37% non-spatial RF accuracy vs. 81.88% spatial Durbin accuracy
2/8
Paper uses a "post hoc" approach to evaluate model performance, where pre-intervention predictions generated by a given model are evaluated against the actual observed post-intervention values rather than a reserved subset of the data (although did that too)
3/8