Need to see whether road safety has improved on a local scale. Lots of macro statistics for London or Midlands but difficult to generate data for precise points.
What it does
On a point by point basis, it find past road works and compares accident datasets before and after the works to display if there has been a change in trend. Future predictions will be done using the Google Prediction API, trained with data from data.gov.uk, to provide forecasts into the future and to predict the effectiveness of planned works to improve safety.
Challenges we ran into
Optimising the model and formatting the data to be accepted by the Google Prediction API. Discovering how to query the Prediction API by location - querying using multiple inputs was not possible.
Accomplishments that we're proud of
Overcame second challenge by hashing location data values into a two dimensional geospacial grid, using the Geohash grid. All points were grouped into 1.5Km squares and stats aggregated.
What we learned
A lot about the Ruby programming language and operational use for ML models.
What's next for Splatter
Getting large data sets which are applicable for training, i.e. for optimising/extending the model.