Weather2Go

Weather2Go

Inspiration:

Our team was inspired by personal experiences driving in Michigan winter weather, where snow and low visibility make even short trips feel risky. After experiencing accidents and insurance rate increases firsthand, we wanted a tool that helps people assess risk before they get on the road.

Function:

Weather2Go asks for your destination, pulls the current local weather, and returns a simple low, medium, or high driving risk level for that area. It also gives a few quick safety tips and general insurance context so drivers can make a more informed decision before heading out.

How We Built It:

We started with a large US accident dataset and focused on Michigan-only data to match local weather conditions. Using live weather data from an API, we trained a Random Forest model to learn how weather patterns relate to accident severity.

To make predictions more reliable, we balanced the data during training and carefully tuned the model to perform well across low, medium, and high-risk levels. This resulted in clear improvements in accuracy and overall prediction quality, showing that weather alone can meaningfully predict driving risk.

We built the web app in Python using Streamlit to connect the model to a working front end, and used Copilot as a learning and troubleshooting tool to help us move quickly and complete the project end to end.

Challenges:

One challenge was finding a dataset with the right weather and accident details, then narrowing a very large dataset down to something usable. We also had to make sure our model was learning correctly by avoiding data leakage and handling imbalanced risk levels.

On the application side, we worked through integration issues connecting our model to a Streamlit app and setting up the live weather API securely. Overcoming these challenges helped us build a more reliable and realistic system.

Accomplishments

We’re proud that we built two independent models on slightly different datasets and got very similar results, which gave us confidence in our approach. We also successfully connected a working front end to our model and live weather API, and completed the project end to end, from data to a usable application.

What's next for Weather2Go

Next, we plan to expand the model with opt-in inputs like driving experience, vehicle type, and trip distance, and add the ability to select a future date and time so drivers can plan trips ahead rather than only assessing current conditions.