The UI displays the hotspots for a given time. The user enters their location...
...and the UI zooms in so they can see the closest hotspots to them.
The results of using a Random Forest Regressor, along with a 3d visualization of the binning & hotspot selection process
An example prediction of hotspots for a random time, and their projection onto a map. Newark airport is an expected notable outlier.
Our (failed) Deep Neural Network model
Uber has become an integral part of our lives over the past few years, but the implications of the ride sharing app, both in terms of environmental and social impact, cannot be overlooked. The core idea behind our project is that Uber data should exhibit characteristic patterns, and is therefore predictable by machine learning algorithms. Being able to predict Uber rides taking place at any point of time is extremely powerful--for instance, with carbon emissions on the rise, this kind of information can be directly helpful in improving gas use efficiency and encouraging ride sharing. Using publicly available Uber data from over from ~4.5 million rides in the Greater NYC area from April-September 2014 (most recent data available in that area), we constructed a machine learning model that predicted the likely ride requests given the time of day and the day of week and leverage it to optimize rider/driver interactions.
What it does
When the RideCast webpage loads, it automatically constructs and trains the machine learning model on the Uber dataset, taking into consideration the current time/day. The UI then displays all of the Uber “hotspots,” which are essentially areas with a very high frequency of rides, in a map. The user enters their location, upon which the map will zoom to their location, allowing them to better view the nearest hotspots to them.
How we built it
Challenges we ran into
Accomplishments that we're proud of
-Training a model to produce an accurate prediction (to within a fraction of a mile) on both latitude and longitude was a difficult, but rewarding challenge. -Completing a real-time interactive visualization of relevant hotspots in your area.
What we learned
-The basics of machine learning in Python with Google’s powerful TensorFlow tools
What's next for RideCast
-Extending it to other datasets than Uber/greater NYC -Making the interactive map more immediately useful by having different search criteria than simply latitude/longitude (currently stands as a proof of concept). -Incorporating directions into it/integrating more with the Uber app itself -Finding local attractions in the area of the hotspots to possibly create a travel app where you can efficiently bounce from venue to venue quickly but with more freedom than public transit.