Inspiration
Economic crises disrupt real households deeply, not just markets.
During COVID-19, some households saved more money and became financially stronger. Others lost jobs, took on more debt, and struggled to stay afloat.
This made us ask:
Can we predict who is financially vulnerable before the next crisis happens?
Does income alone truly determine financial resilience — or do savings, debt, home ownership, education, and location matter just as much?
What it does
Our model predicts whether a household is likely to:
- Improve
- Stay the Same
- Worsen
during a future economic shock similar to COVID-19.
Instead of giving just one answer, our model estimates:
$$ P(\text{Improved}) \quad P(\text{Stayed Same}) \quad P(\text{Worsened}) $$
This provides a clearer picture of financial risk and resilience.
We also built an optional Streamlit demo where users can enter a financial profile and instantly see predicted probabilities.
How we built it
We used the Survey of Financial Security (SFS) dataset, which contains detailed demographic and financial information about Canadian households.
Our approach:
- Cleaned and mapped survey data
- Handled missing values
- Selected key financial and demographic features
- Preprocessed data for modeling
- Trained multiclass models (Logistic Regression, Random Forest)
- Evaluated using accuracy, F1-score, and confusion matrices
- Identified key predictors using feature importance
Challenges we ran into
- Managing class imbalance across outcome categories
- Balancing predictive performance with interpretability
- Ensuring reproducibility within limited hackathon time
Accomplishments that we're proud of
- Successfully using Logistic Regression for the first time
- Creating a clean, reproducible workflow
- Developing an interactive demo for easier understanding
What we learned
- How to work efficiently with a large dataset in limited time
- The importance of selecting meaningful features
- The value of clear communication alongside technical work
Log in or sign up for Devpost to join the conversation.