Road Accidents due to Climate Change

We want to show the world how changes to their climate will affect them in their day-to-day life: in terms of risks in driving through their local neighborhoods

Comment

Inspiration

We are inspired to communicate to the public the real-world consequences of Climate Change & how it will effect them every day: in terms of risks on their residential roads.

What it does

Show how climate change will affect something they interact with every day: road conditions Provided: ClimateData.ca contains historical and future climate projections from multiple models

How we built it

Jupyter notebooks for data extraction, analysis, and algorithmic training

Challenges we ran into

Interpretation of our features from climatedata.ca – Difficult User-Friendly interpretation. Computational Expensive Regressor Models – Gradient Boosting Forests Associated Accident data into the correct grid – formatting and labelling accidents to their respective grids.

Accomplishments & what we learned

Implementation of forest-based Regressors – Usually implementation around Classification problems for forest-based models. Implementation of DNN Neural Networks. Climate Change forecasts in the future – how impactful Climate change will be in densely populated sectors in Montreal – suggestions towards pushing less densely populated regions to populated regions (Flooding as an example we are currently seeing on the island of Montreal).

What's next for Road Accidents due to Climate Change

Continue tuning these algorithms Expand the number of grids considered to other locations (all of Canada?) Investigate CNNs on the grids Construct a report alerting the public of the dangers of Climate Change and how it will affect driving conditions

Contents/Glossary

Clean Data.ipynb

  • Clean precipitation data to be merged later to the accident data

Create Future Validation Set.ipynb

  • Create future set to be used to forecast accident data from 2018-2100

Linear Regression.ipynb

  • Train Linear Regression algorithm on data as baseline

Merge Data.ipynb

  • Merge the accidents and precipitation datasets to be sent to training

Mila2019ClimateChangeHackathon_DNN.ipynb

  • Training of a Dense Neural Network on predicting # of incidents based on location, time, and precipitation

Summary_Results_CC

  • Jupyter notebook depicting the average results for each of the possibly target variables. Code includes a table that outlines the averages for targt variables from the testing set coming from the test-train set.

Forests_ClimateChange

  • Random Forest and Gradient Boosting Forest Regressors to predict the output of Number of Incidents – Notebook includes the test-train splits, the GridSearch hyper tuning parameters and the Mean Absolute Error for the random forest and gradient boosting forest results.

RF_Script

  • Random Forest script – for looping all the possible target values and outputting in dataframe the regressor scores ( Mean Absolute Error / Average for each possible target variable).

Built With

Share this project:

Updates