For a narrative summary and key insights on this project, please read my blog post Predict Customer Churn With Precision on Medium.
Telco Co. wants to deploy customer retention strategies across systems and business processes to reduce customer churn which is currently running as 26.7% of customers. Business requirements include:
- Find the best initial prediction model to classify customer churn risk
- Model performance should do significantly better than using "averages"
- Deliverables should explain the relative influence that each predictor has on the overall model predictions
- Deliverables should suggest potential solutions to reducing customer churn
The project involved analyzing, cleansing, plotting, featurizing and modeling about 7K customers historical data that the churned or were retained. After analyzing and transforming the data, I optimized several classification models using GridSearchCV. Each model was trained on 80% of the historical data and then asked to predict churn scores on the remaining 20% test data. I used the tools and libraries from Python, Numpy, Pandas, Seaborn, StatsModel, and Scikit Learn. Here you can find my final business presentation and my Jupyter notebook.
Below is a list of the contents of this repository - instructions for using them are in the next section.
- README.md: The README for this repo branch explaining it's contents - you're reading it now.
- Telco-Churn-Classification-Model.ipynb: Jupyter Notebook containing background, data sources, scripts, data profiling, cleansing, analysis, feature engineering and models. Extensive comments included.
- Customer_Churn_Classification.pdf: Final business presentation targeting a 5-minute summary pitch
- Tableau Workbooks: Several Tableau workbooks were used for exploratory data analysis and visualization
- data folder: Folder contains source data files available for use in the project
- images folder: A folder for some images of plots used in presentation
- .gitignore: A hidden file that tells git to not track certain files and folders