Richard Wen, Songnian Li
rrwen.dev@gmail.com, snli@ryerson.ca
The purpose of this research is to develop an approach for a Spatial Decision Support System (SDSS) that integrates Geographic Information Systems (GIS), Automated Machine Learning (AutoML), and Hyperparameter Optimization (HPO) to generate precision geo-interventions based on standardized geospatial data and user design constraints. The geo-intervention generation approach involves three steps: (1) Geo-binning, (2) AutoML, and (3) Prediction Optimization. Geo-binning is used to standardize geospatial data into regularized grids as inputs into AutoML models. Prediction optimization generates geo-interventions by applying user-design constraints and optimizing AutoML model output to find optimized input variables that form precise geo-interventions. An experiment in reducing road traffic collisions using infrastructural changes in Toronto, Ontario, Canada was done to evaluate the geo-intervention generation approach. The results of the experiment found that changing the number of schools, red light cameras, and transit shelters in high traffic areas could potentially halve the total number of traffic collisions according to a 80 by 80 geo-binned grid Auto-Sklearn model with a Mean Absolute Error (MAE) of 117.68. It was also found that user design constraints heavily affected the prediction optimization step as when the areas were altered to an alternative grid of cells with scarce infrastructure, the number of predicted collisions rose by 6127 collisions. Thus, limitations of this study included subjectivity in user design constraints, scalability, and interactivity. Future work involves improving modelling/optimization efficiency and developing an interactive interface for exploring generated precision geo-interventions.
Date | Note |
---|---|
Wednesday, November 2, 2022 8:30-10am (Beijing TZ) | Presented slides (12 min, 3 min questions) for Urban Geo-Info Session (Link) |
Thursday, October 27, 2022 | Full paper published to ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences (PDF | Web) |
Tuesday, October 23, 2022 | Presentation slides submitted to Urban Geo-Information track for November 2, 2022 Urban Geo-Info Session at 8:30-10am (PDF | PPT) |
Tuesday, September 20, 2022 | Full paper submitted to the Urban Geo-Information track (PDF | DOC) |
Thursday, September 1, 2022 | Abstract accepted to the Urban Geo-Inforomation track (PDF | DOC) |
Thursday, August 18, 2022 | Abstract submitted to the Urban Geo-Information track (PDF | DOC) |
In Windows:
- Install Anaconda3
- Install Windows Subsystem
wsl --install
- Install dos2unix
wsl sudo apt install dos2unix
- Note: You may be promoted for a password due to
sudo
privileges
- Note: You may be promoted for a password due to
- Run
bin/setup.bat
to convert scripts inbin
folder for windows use - Enter a Windows Subsystem terminal with
wsl
- Run
bin/install
to create aconda
environment - Activate the
conda
environment (namedgi4dm-2022-paper
)
wsl --install
wsl sudo apt install dos2unix
bin\setup
wsl
source bin/install.sh
source bin/activate.sh
In Linux/Mac OS:
- Install Anaconda3
- Run
bin/install
to create aconda
environment - Activate the
conda
environment (namedgi4dm-2022-paper
)
source bin/install.sh
source bin/activate.sh
Richard Wen rwen@ryerson.ca and Songnian Li snli@ryerson.ca