DEV Community

Erik Anderson
Erik Anderson

Posted on

Uber Surge Predicted with Machine Learning | Chicago

https://youtu.be/M016woop34c
For starters, there's a youtube video, or you can just read below, if that's more your speed.

Rideshare Helper

https://github.com/ekand/rideshare_helper

What can rideshare drivers in Chicago do to earn more money?

This is a new thing: They can use this data science project with its machine learning model which predicts when and where high demand pricing will occur.

High demand pricing goes by different names at different companies: At Uber, it is called Surge Pricing, at Lyft, it's called Prime Time Pricing.

For this project, I used data from the city of Chicago: A dataset of rideshare trips (https://data.cityofchicago.org/Transportation/Transportation-Network-Providers-Trips/m6dm-c72p), a data set of percapita income, among other health factors (https://data.cityofchicago.org/Health-Human-Services/Per-Capita-Income/r6ad-wvtk), and data set with geographical outlines for the 77 community areas in Chicago (https://data.cityofchicago.org/Facilities-Geographic-Boundaries/Boundaries-Community-Areas-current-/cauq-8yn6). The rideshare dataset contained about 129 million trips over a bit more than a year, but I focused on one arbitrarily selected day to do my analysis. This greatly simplified things and allowed me to perform the analysis on my laptop.

The techniques used were Linear Regression for feature engineering and Random Forest Classification to predict whether a given trip will be under surge pricing.

The Random Forest Classifier Model performed passably well: It had 25% precision, although the recall was quite low. However, this is adequate for my use case: If I can tell drivers "go to this place at this time and you've got a 25% chance of catching a surge," I think that's valuable.

The model also idendified important features: proximity to the loop, per capita income in the community area where the trip starts, and whether the trip start between 4pm and 8pm. In other words, high demand pricing is more likely to occur downtown
This is a nice sanity check to ensure the model makes sense.

In the PDF version of the presentation, you can see some visualizations of the model's predictions. There are also live, interactive visuzlization here (https://public.tableau.com/profile/erik.kristofer.anderson#!/vizhome/RideshareHighDemandPricingPredictions/predict_morn_eve?publish=yes) and here (https://public.tableau.com/profile/erik.kristofer.anderson#!/vizhome/RideshareHighDemandPricingPredictions/true_and_pred_dashboard?publish=yes)

The jupyter notebooks contain most of the code, and the detailed analysis.

The organization of the project is as follows:

------------
.
├── data              -----> contains various data files, organized and excluded from version control 
│   ├── external
│   ├── interim
│   ├── pickles
│   ├── processed
│   └── raw
├── docs             -----> documentation (currently empty)
├── models           -----> pickled models
├── notebooks        -----> jupyter notebooks with main code and explanation
├── references       -----> external reference, if needed
├── reports          -----> presentation files
│   ├── external_images
│   ├── figures
│   └── videos
└── src               -----> Python scripts, packaged as modules and imported using `pip install -e .`
    ├── data_prep
    ├── features
    ├── models
    ├── pickle
    ├── surge
    └── visualization

--------
Enter fullscreen mode Exit fullscreen mode

Project based on the cookiecutter data science project template.

"https://drivendata.github.io/cookiecutter-data-science/"

#cookiecutterdatascience

Hope you enjoyed reading this and please post below if you have questions (or if you just liked it, or even if you didnt'; I just love comments).

Discussion (0)