🔧 Keywords

Forecasting, Feature Engineering, Regression Models, Neural Networks, Ensemble Methods, Scikit-learn, MLP, EDA, XGBoost

🧩 Problem

Accurate sales forecasting helps companies plan inventory, allocate resources, and reduce overstock or missed demand. The goal of this project was to predict daily product sales using historical data and tabular features (e.g. product type, price, store, date).

This type of forecasting is critical for both supply chain optimization and business decision-making.

⚙️ What I Did

• Explored and cleaned the sales dataset: handled nulls, outliers, date formatting, and missing values.
• Engineered features including product category, store ID, day-of-week, promotions, and price.
• Built and compared multiple forecasting models:
  • Linear Regression
  • Multilayer Perceptron (MLP) for tabular prediction
  • Ensemble models (Random Forest, XGBoost)
• Tuned hyperparameters using grid search and cross-validation.
• Evaluated models with RMSE, R², and MAE, and visualized prediction vs. actuals.

📊 Outcome

• Identified the best-performing model (XGBoost) with improved accuracy over baseline.
• Highlighted feature importance to explain key sales drivers.
• Demonstrated practical ML forecasting skills on real-world business data.

🛠️ Tech Stack: