Skip to content
Case Study for ride sharing service regaring User Churn
Python
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
images
.gitignore
README.md
definitions.py
model.py

README.md

User Churn Case Study

This case study examines user data of a ride share company and attempts to predict when a user is at risk of churning. The following describes the structure of this repository and outlines some of the findings. A power point presentation on this study can be found here. The data is ommited from the repository as per request of the company.

Structure:

  • Initial EDA was done in eda.ipynb file
  • The bulk of the model including data cleaning is done in churn_model.py
  • Some helper functions for generating profit curves in in helpers.py
  • Images directory contains
    • Profit curves with different budgets
    • Feature importance plot
    • ROC plot of our model

Model:

  • The final model used was a Gradient Boosted Model
  • It performed slightly better than a Random Forrest and significantly better than a Logistic Regression model
  • A Grid Search was run to find the optimized parameters
  • Final Model had an AUC score of .845

Findings:

  • Several different budgets were created assuming that one way to get an at-risk customer to not churn was a promotion
    • We created several different budgets with this assumption in order to figure out what the optimal threshold would be in considering a user at-risk
    • These findings are displayed in the images folder
    • Analyzing the feature importance in our Gradient Boosted model revealed the the strongest factors were:
      • The % of time the user rode during surge pricing
      • The % of time the user used the service during the weekday
You can’t perform that action at this time.