🚀 Project Objective Predict weekly units sold for each SKU-store combination to support more accurate demand planning and inventory decisions.
This repository contains the full code, process, and learnings from a demand planning project where I tackled data preparation, feature engineering, and model evaluation using both linear and tree-based approaches.
🔍 Key Techniques Used 🧱 Feature Engineering Log and power transformations to normalize highly skewed sales and pricing data.
Creation of aggregated features like:
Average units sold by SKU, store, and time period.
Interactions between store_id and sku_id.
Date-based features (e.g., month, weekday, week_of_year).
📈 Modeling Baseline models: Linear Regression, Ridge, Lasso
Gradient Boosting models: XGBoost and LightGBM
Comparison of performance using MAE, RMSE, and R² metrics