Back to projects

Predictive ML / Explainability

Customer Churn Prediction & Risk Dashboard

A supervised churn prediction system that turns customer records into risk scores, SHAP-backed explanations, and dashboard-ready decisions.

Live DemoGitHub
Main result0.88ROC-AUC
accuracy0.86
precision0.74
recall0.58
F1-score0.65
Interactive Streamlit dashboard demo, shown with native video controls.
Problem

Customer churn risk is hard to prioritize when teams only see raw customer records and disconnected business signals.

Solution

Built an end-to-end ML workflow with preprocessing, engineered features, model comparison, explainability views, and a Streamlit dashboard.

Outcome

The final system made churn risk easier to inspect, compare, and explain through dashboard workflows instead of raw tables.

Workflow

01Clean data
02Engineer features
03Train models
04Explain risk
05Dashboard

Role and technical focus

ML pipeline, feature engineering, evaluation, dashboard UI

PythonScikit-learnXGBoostSHAPStreamlitPandas

Notes

  • Weighted ensemble using Random Forest and XGBoost.
  • Explainability flow includes SHAP and partial dependence style views.
  • Dashboard supports risk calculator and probability visualization workflows.
Project media

Real screens and analysis visuals