top of page

Telco
Customer Churn

Data Science Weekend (DSW) 2023 is an event held by Data Science Indonesia bringing the theme Nurturing Data AI and Digital Talent for Tomorrow. One of the series of events, Data Challenge, is a competition that aims to analyze data and provide solutions to given problems. Access to the dataset remains private and any distribution is prohibited.

Competition

Data Science Week Indonesia 2023

​

Final Standing (Grouping)

1st Runner-Up (Rank 2/167 teams)

​

Project Type

Visualisation, Classification, Business Insights, Business Strategy

​

Date

November 2023

Abstract Sphere
image.png

Data Visualisation

Data visualization was instrumental in revealing key patterns and relationships. Through vibrant charts, we decoded how features like tenure months, device class, etc. contribute to the prediction of customer churn and its effect on revenue gain or loss. These visual insights not only enriched our understanding but also guided our modeling decisions, transforming raw data into an enlightening narrative about customer churning issues.

Predictive Modelling

Predictions are performed using various ML algorithms such as Logistic Regression, Support Vector Machine, and XGBoost. In the end, Voting Classifier would aggregate and decide on the final outcome of whether a particular customer is likely or unlikely to churn. Each model has its own pros and cons, and deciding which model to rely on would depend on whether we prioritised predictive power or interpretability of our result.

image.png
image.png
image.png

Model Evaluation & Interpretability

It's not entirely a black box! The model underwent rigorous evaluation, ensuring its predictions were not just accurate but also consistent across varied data subsets. By employing tools like SHAP, ICE, and Cumulative Gain Charts, we attempt to make a decision-transparent XGBoost model, bridging the gap between high-dimensional data complexities and actionable business insights.

Andreas Lukita - 2nd Winner Student & Junior Professional-1.png

Check out my work

For a more immersive experience, check out my Streamlit App.

For code details, check out my GitHub page

  • LinkedIn
  • Medium
  • Twitter

© 2023 by Andreas Lukita.

bottom of page