In-N-Out Location Predictor

• 

  Model Heatmap For Irvine

• 

  LightGBM Feature Importance

• 

  Omni Dashboard

• 

  EDA, performed with OMNI and love

• 

  More Detailed EDA

• 

  A phone call with In-N-Out in the pursuit of revenue data

Presentation

https://www.figma.com/deck/SIYvTOEJ6oBbFXRnUYAVVI

Inspiration

We knew of an empty lot at an intersection near our house and we were curious what would be built there, hoping it to be an In-N-Out. This got us thinking, what makes for a "good" In-N-Out location, why do they choose to expand to some locations and not others.

What it does

This project determines the similarity between any coordinate point location and a previously established In-N-Out, helping determine feasibility for a location to be opened.

How we built it

We constructed our dataset with various public sources. Caltrans Performance Measurement System, Road Distance API, OpenStreetMap, U.S. Census. We collected features like distance to other In-N-Out's, distance to distribution centers, median income, distance to freeway ramp, Annual Average Daily Traffic, daytime population, etc.

After collecting features, we trained an XGBoost model as well as LightGBM and compared results to determine which model would yield more useful predictions.

Challenges we ran into

Challenges within the features included erroneous values which had to be removed, such as a Chipotle claiming to be over 600km away from the nearest In-N-Out. We also tried getting revenue data which was not publicly available in order to determine a store's success. We called In-N-Out, but they were unable to provide us with this data. As for the models, their accuracy was exceptional although we found that XGBoost especially would return high likelihoods in clearly wrong places like the middle of the ocean. This was fixed by filtering for only selecting locations with a surrounding population.

Accomplishments that we're proud of

We are proud of joining multiple data sets from various domains cleanly and effectively. We are also very pleased with how we learned and used Omni to create visualizations which helped us tell a clear story with our data, bringing our project from technical abstractness to concrete ideas.

What we learned

We learned how to apply geospatial data in training models, Omni dashboards, combining multiple datasets. Beyond data, we learned about decision-making in the fast food industry, what are large food corporations looking for in the pursuit of expansion.

What's next for In-N-Out Location Predictor

Our proposal is to make an interactive web app beyond just In-N-Out. Generalizing the model to any fast food chain and allowing users to see possible expansion locations on their own map.

Built With

• lightgbm
• python
• xgboost