How ahmed-m-fayad/customer-churn-prediction-and-analysis works
Project Type and Purpose
• This is an end-to-end machine-learning pipeline for customer-churn prediction.
• It comprises three sequential ETL/ML stages:
a. Data Collection & Preprocessing
b. Advanced Data Analysis & Feature Engineering
c. Model Development & Optimization (including final “complete” deployment pipeline).
Main Components & Pipelines
• Raw Data Store
• test.csv, train.csv (in Data Collection/Raw Data)
• Cleaning Pipeline
• cleaning_pipeline.joblib transforms raw → cleaned & imputed CSVs
• Outputs: train_cleaned.csv, train_cleaned_imputed.csv, train_split_cleaned(.csv), validation_split_cleaned.csv
• Feature Engineering Pipeline
• featureEng_encoder_scaler_pipeline.joblib encodes/scales cleaned splits
• Outputs: train_basicFeatureEng.csv, train_split_preprocessed.parquet
• Model Training Pipeline
• complete_churn_pipeline.pkl wraps full preprocessing + model
• custom_transformers.py, feature_pipeline.py define transformations
• best_churn_model_logistic_regression.pkl & xgb_clf_all_features.joblib are final models
• Reports & Notebooks
• Jupyter notebooks record exploratory steps, milestone docs generate PDF/Word reports
Relationships & Data Flow
• Cleaned Data Store → Feature Engineering Pipeline → Feature Store
• Feature Store → Model Training Pipeline → Model Artifacts & Deployment Pipeline
Architectural Patterns & Design Principles
• Sequential, modular ETL/ML stages (each persists artifacts in joblib/parquet/pkl files)
• Reusability: each pipeline exists as a serialized artifact (joblib/pkl) that can be re-invoked
• Clear separation: data ingestion, transformation, modeling, and reporting
Technologies & Frameworks
• Python, pandas, scikit-learn, XGBoost
• joblib for pipeline serialization, parquet for efficient storage
• Jupyter for notebooks, PDF/DOCX for reports
– Input: Raw Data
– Output: cleaned CSVs (list them)
– Input: cleaned splits
– Output: preprocessed parquet & basic feature CSV
– Input: preprocessed data
– Output: final models (.pkl, .joblib)
f. Surrounding: Reports & Notebooks cluster off to the side, linked back to each stage