How true-star-580/dsc_phase_3_project works
Project Type and Purpose
• This is a data‐science/ML pipeline for customer churn prediction. It ingests raw customer data (CSV), performs EDA and preprocessing in Jupyter, trains and evaluates multiple classifiers, selects an optimized XGBoost model, serializes it to disk, and produces outputs (model file, presentations).
• The goal is to predict individual churn risk and enable targeted retention workflows.
Main Components
a. Data Source
• bigml_59c28831336c6604c800002a.csv (raw customer records)
b. Data Processing & EDA
• index.ipynb notebook: exploratory analysis, cleaning, feature engineering, imbalance handling (SMOTE)
c. Model Training & Evaluation
• index.ipynb: dummy, logistic, decision tree, GBM, XGBoost models; cross‐validation; hyperparameter tuning
d. Model Artifacts
• xgb_model.pkl (serialized XGBoost model)
e. Reporting & Presentation
• Non-technical Presentation.pdf
• Presentation_Paul Kamau.pdf
• README.md documentation
Relationships and Data Flow
• Raw CSV → EDA Notebook → Cleaned/Engineered Dataset → Model Training → Evaluation Metrics → Best Model Selection → Serialization (xgb_model.pkl) → Reporting Dashboards/Presentations.
Architectural Patterns & Principles
• Modular Pipeline: clear separation of ingestion, cleaning, modeling, and reporting stages.
• Reproducibility: Use notebooks to capture analysis steps; serialize model for deployment.
• Artifact Management: CSV and pickle file are versioned inputs/outputs.
• Scalability Concern (future): swap CSV for data warehouse or streaming source; wrap notebook steps in scripts or workflow orchestrator.
Technologies & Libraries
• Jupyter Notebook (index.ipynb) for orchestration of steps
• Python data stack: pandas, numpy, scikit-learn, imbalanced‐learn (SMOTE)
• XGBoost library for final model
• Pickle for model persistence
• Reporting via PDF exports or future dashboard (e.g., Streamlit, Dash)