Skip to content

README.md

Latest commit

 

History

History
120 lines (84 loc) · 3.33 KB

README.md

File metadata and controls

120 lines (84 loc) · 3.33 KB

🏪 Walmart Weekly Sales Prediction

CDSD Certification Project — Linear & Regularized Regression

📋 Executive Summary (click to expand)

Objective: Predict weekly sales for 45 Walmart stores to optimize inventory, marketing campaigns, and minimize overfitting.

Target KPI: R² ≥ 90% on unseen data

Dataset:

  • 6,435 weekly records, 45 stores, 7 features + temporal variables
  • Target: Weekly_Sales ($)
  • Preprocessing: outlier removal (Z-score 3σ), temporal feature engineering, 5,912 clean rows, 80 features

Pipeline Highlights:

  • ColumnTransformer + GridSearchCV
  • Numerical: KNNImputer → StandardScaler
  • Categorical: OneHotEncoder (handle_unknown='ignore')
  • Target leakage fully prevented

Models Evaluated: Linear Regression, Ridge (α=0.01), Lasso (α=500)
Validation: Train/Test split + 5-fold CV

🔬 Model Evaluation & Results
Model R² Train R² Test Overfit RMSE MAE
Linear Regression 0.9714 0.9640 0.0074 130,948 103,671
Ridge (α=0.01) 0.9713 0.9630 0.0083 132,698 104,789
Lasso (α=500) 0.9708 0.9634 0.0073 131,977 102,517

Chosen model: Lasso Regression

  • Excellent predictive performance
  • Minimal overfitting
  • Sparse coefficients (~60% zeroed)
  • Improved interpretability for business stakeholders
📊 Key Business Insights
Insight Impact Recommended Action
Store dominance Top 10 stores = 45% total sales Focus inventory on high performers
Holiday effect +22% sales Pre-stock 2–3 weeks before holidays
Economic sensitivity Sales negatively correlated with unemployment Adjust promotions during downturns
Seasonality Nov/Dec peaks Plan staffing & marketing campaigns

💰 Estimated annual business impact: ~$120M (forecast accuracy + inventory & holiday optimization)

🛠️ Production-Ready Pipeline
  • ColumnTransformer + GridSearchCV
  • Pipeline export: preprocessor.pkl, lasso_model.pkl
  • FastAPI endpoint: POST /predict_sales → store-specific weekly forecast
  • Docker / AWS Lambda ready (<100ms inference)
  • Drift monitoring: retrain automatically if R² < 90%
✅ CDSD Certification Coverage
  • EDA & preprocessing
  • Linear regression baseline
  • Regularized models (Ridge & Lasso)
  • Cross-validation & overfitting control
  • Feature importance & business interpretation
  • Production-ready ML pipeline & deployment artifacts
🚀 Quick Start 
# Clone the repository
git clone https://github.com/Data-Science-Designer-and-Developer/Project_Walmart.git
cd Project_Walmart

# Install dependencies
pip install -r requirements.txt

# Run the notebook
jupyter notebook
<<<<<<< HEAD
  1. Run the notebook sequentially
  2. Use deploy_pipeline.py to generate production artifacts (.pkl)
  3. Use predict.py to forecast store sales

👨‍💻 Author

Dreipfelt — CDSD Data Science Certification Candidate GitHub: https://github.com/Dreipfelt