This project forecasts daily sales for Rossmann stores using historical data, store metadata, and engineered features. We use the XGBoost regression model to capture complex patterns and improve predictive accuracy.

• Train Data: train.csv
• Test Data: test.csv
• Store Metadata: store.csv

Data Source: Rossmann Kaggle Competition

• Predict the Sales column using historical store data.
• Create a model that generalizes well to unseen data (test set).
• Submit predictions in submission.csv.

• Python
• Pandas, NumPy, Matplotlib, Seaborn
• XGBoost
• Scikit-learn for model selection and metrics

• Store, DayOfWeek, Promo, SchoolHoliday, StateHoliday, StoreType, Assortment, CompetitionDistance, Promo2

• Year, Month, Day, WeekOfYear

• IsPromoMonth — True if the current month is a promo month for that store
• IsHolidayWeek — True if a holiday (school/state) occurred in that week
• CompetitionOpenTimeMonths — Number of months since a competitor opened
• Promo2OpenTimeWeeks — Weeks since Promo2 started

• Sales_Lag1, Sales_Lag7, Sales_RollingMean3, Sales_RollingMean7

🚫 These features are excluded from test data since future sales are unknown.

1. Load & Merge: Combine train.csv/test.csv with store.csv
2. Feature Engineering: Extract and transform useful features
3. Train/Test Split: Split train data for model evaluation
4. Train Model: Use XGBRegressor
5. Hyperparameter Tuning: GridSearchCV (optional)
6. Predict: Apply best model to processed test data
7. Submit: Create submission.csv

• Actual vs Predicted Sales (scatter plot)
• Predicted Sales Trends (by Date or Store)
• Feature Importance from XGBoost