How darshankamate1254/customer-churn-prediction-system- works
This repository is a customer-churn ML pipeline: it cleans and splits Telco data, engineers features, trains and logs three classifiers, and exposes saved-model scoring through a Streamlit UI. The UI supports individual and CSV batch scoring, risk recommendations, and downloadable results; evaluation utilities generate classification plots and SHAP analysis. Training-to-preprocessing and training-to-feature-pipeline calls are visible in source. The README documents EDA, MLflow, and app capabilities; unsampled evaluation or notebook wiring is not inferred.
Main entry points: train.py Model training · app.py Scoring UI · README.md Saved models · preprocess.py Data cleaning
Components 15
Data preparation
- Telco dataset
data/raw/README.md - Data cleaning
src/preprocess.py - Train/test split
src/preprocess.py - Exploratory analysis
notebooks/01_eda.ipynb
Modeling and evaluation
- Feature pipeline
src/features.py - Model training
src/train.py - Model evaluation
src/evaluate.py
Scoring experience
- Scoring UI
app/app.py - Single scoring
app/app.py - Batch scoring
app/app.py - Risk recommendations
app/app.py - Score results
app/app.py
Artifacts and tracking
- Saved models
models/README.md - MLflow tracking
Other components
- Analyst
Connections 19
- Analyst → Exploratory analysis: explores data
README.md - Analyst → Model training: starts training
README.md - Telco dataset → Data cleaning: loads and cleans
- Model training → Data cleaning: calls cleaning
src/train.py - Model training → Train/test split: calls splitting
src/train.py - Model training → Feature pipeline: builds pipeline
src/train.py - Data cleaning → Train/test split: returns cleaned data
src/preprocess.py - Analyst → Scoring UI: opens app
README.md - Scoring UI → Saved models: loads model
app/app.py - Scoring UI → Single scoring: selects mode
app/app.py - Scoring UI → Batch scoring: selects mode
app/app.py - Single scoring → Saved models: predicts probability
app/app.py - Batch scoring → Saved models: scores customers
app/app.py - Single scoring → Risk recommendations: assigns risk actions
app/app.py - Single scoring → Score results: displays gauge
app/app.py - Batch scoring → Score results: exports scores
app/app.py - Model training → Saved models: saves pipelines
src/train.py - Model training → MLflow tracking: logs runs and models
src/train.py - Model evaluation → MLflow tracking: logs plot artifacts
src/evaluate.py