GBDi - Accidents in Thailand

• 

  To later be deployed into Azure WebApplication using Dockerfile

Inspiration: It is a problem common across all countries especially South East Asia. A simple yet effective dashboard on trending of accidents reported historically could help government officials or authorities take quick action.

What it does: Live dashboards and applications built with Voilà and Jupyter widgets on accidents in Thailand

How we built it:

1. ipywidgets and voila for this MVP
2. (Intended) Deploy in web application in Azure but faced some limitation so Voila local is used (Azure account is elaine.khoosynnyie@gmail.com)

Challenges we ran into:

Challenges and Limitations:

1. Model Deployment: This dashboard faced some limitation deploying into web application but it is doable.
2. Model Performances: Models above can be further fine tuned using different minimization methods as well as seasonality and autoregressive parameters
3. Feature Engineering: Rather than using time series modeling, new features (external factors like humidity in that province, initiatives by government officials to bring down accident rates, daily traffic in that province etc) can be added into the dataset and be modeled as a multivariate time series model or regression problem. Note: There was an attempt done to model it through linear regression where features are generated based on lagging of its own features but performance of the model is worse than the above.

Accomplishments that we're proud of:

1. Interactive dashboard for quick visualization of accident rates and population at risk
2. Preliminary modeling using exponential smoothing to predict next N days.

What we learned:

1. The need for intervention from government or authorities is necessary to bring down the cases of accidents in Thailand especially central Thailand.

What's next for GBDi - Accidents in Thailand:

1. Deployment of dashboard (descriptive and predictive)
2. Different modeling techniques (multivariate time series or regression)
3. Clustering algorithm to find out if there're common places that have similar accident rates.

Built With

• ipywidgets
• jupyter
• notebook
• voila