The goal of the challenge is essentially to recommend a Driver's Lifetime Value (i.e., the value of a driver to Lyft over the entire projected lifetime of a driver). This notebook details the solution to Lyft Data Science Challenge.
Dataset was emailed to all participants by Lyft. We are provided with driver_ids.csv
, ride_ids.csv
, and ride_timestamps.csv
.
All work was done on Google Colab. To reproduce the result, first create a folder under Colab Notebooks. Otherwise, remove the cells that involve Google Colab and change the following path locally.
PATH='/content/drive/My Drive/Colab Notebooks/lyft'
To install the needed libraries locally, run pip install -r requirements.txt
LTV is affected by the number of days the driver drove, total revenue generated by the driver, which is related to miles traveled, minutes traveled, percentage increase in the fare, and the churn rate.
The average projected lifetime of a driver is 3.03 years.
Drivers can be segmented into 3 main clusters by applying KMeans clustering algorithm to features associated with LTV. We also study other features based on clusters. Results show that there are several features that are significantly different based on each group. For instance, the number of rides completed on weekdays and time elapsed between a driver's drop-off and pick-up time.
We recommend that high value drivers should be rewarded with higher share of fare; mid value drivers should be nudged to encourage driving more consistently based on unusual inactivity; low value drivers should be incentivized via gamification strategies. Loyalty program can be introduced to all drivers to discourage "dual apping".
Thanks to Lyft for organizing this competition and providing an interesting dataset for analysis.