Our inspiration for this technology is increasing public accessibility to the insights derived from machine learning data to advance the greater health of our community. With the resources, intellect and data at Statefarm's disposal, we can harness valuable insights and create excellent UIs to increase awareness of this data.

What it does

Our app uses a machine learning model derived from the Statefarm-supplied crash data to analyze real time weather data streaming from a user's device and inform them of the possible severity of a crash. It then displays factors to watch out for in order to decrease that severity if the severity met a certain threshold.

How we built it

Technologies we used include:

  • Tensorflow, for creating the machine-learning model
  • Flask, for making an API to access data from the model
  • React-native, for cross-platform mobile development
  •, for easy prototyping and displaying of app functionality
  • Open Weather API, for accessing real-time weather data

Challenges we ran into

  • Handling such a large csv file on our local machines and performing analyses with our computers was time-consuming and patience-demanding ;)
  • It was hard to create a properly-functioning splash screen in React-native.
  • Cross-compatibility in running the model between our Windows and Mac machines was an issue at times.
  • Implementing push notifications was a challenge we unfortunately could not overcome.

Accomplishments that we're proud of

  • A clean, robust cross-platform UI that handles a wide range of devices.
  • Our first implementation of machine learning to predict outcomes based on known variables.
  • Successful interaction between a Python-based API and a Javascript-based front-end, as well as implementation of JS-specific APIs.
  • Beautiful weather-based background gradients.

What I learned

  • is an amazing tool for creating quick react-native applications and displaying them on a phone.
  • React-native is more forgiving than Flutter, XCode for programmers new to front-end.
  • Every machine learning approach has its ups and downs, and time should be taken to decide which strategy is best for analyzing a set of data.
  • 2 hours of sleep is more than enough for the final Hackathon rush ;)

What's next for Like a Good Neighbor?

  • Background GPS updating with push notifications (enable off-app support).
  • Test different learning models.
  • Use frequency of crashes in certain situations in addition to using severity for a more robust model.
  • Do further testing of accuracy of the model.
  • Enable visualization of data for the end-user.

Built With

Share this project: