🏪 Walmart Weekly Sales Forecasting

📘 Project Overview

This project aims to forecast weekly sales for Walmart stores using machine learning techniques.
By analyzing historical sales data with store, department, and holiday information, we build models that predict future weekly sales with improved accuracy.

---

📂 Dataset Description

1. Train Dataset

• Contains historical weekly sales records.
• Columns:
  • Store: Store number
  • Dept: Department number
  • Date: Week date
  • IsHoliday: Whether the week contains a special holiday
  • Weekly_Sales: Target variable (total sales for that week)

2. Features Extracted

After preprocessing and feature engineering:

• Lag features: lag_1, lag_2, lag_3, lag_4, lag_52
• Rolling mean features: rolling_mean_4, rolling_mean_12
• Temporal features: Year, Month, Week, Day
• Holiday feature: IsHoliday (encoded)
• Scaled features using MinMaxScaler

3. Test Dataset

• Includes same features except Weekly_Sales (to be predicted).

---

⚙️ Project Workflow

1. Data Preprocessing

• Loaded train and test data.
• Converted Date column to datetime.
• Sorted data by Store, Dept, Date.
• Merged datasets for consistent feature generation.
• Created lag and rolling mean features to capture past sales trends.

2. Feature Engineering

• Extracted time-based features (Month, Week, Year).
• Added historical sales behavior using lag & rolling averages.
• Scaled numerical features using MinMaxScaler.

3. Data Splitting

• Split train dataset into:
  • Train (70%)
  • Validation (15%)
  • Test (15%)

---

🤖 Model Training

Trained multiple regression models to predict Weekly_Sales.

Model	Description	Status
Linear Regression	Baseline model	✅
Random Forest Regressor	Ensemble tree-based model	✅
Decision Tree Regressor	Simple tree model	✅
Support Vector Machine (RBF)	Kernel-based regression	✅
XGBoost Regressor	Gradient boosting model	✅

---

🔍 Model Evaluation

Metrics Used:

• R² Score
• Mean Absolute Error (MAE)
• Root Mean Squared Error (RMSE)

Evaluation Workflow:

1. Trained models on training data.
2. Predicted on validation and test sets.
3. Compared all models’ performance.

---

📊 Visualization

Actual vs Predicted Plot

Shows how closely the model predictions align with actual sales trends over time.

💾 Model Saving

All trained models and scaler were serialized using pickle for future predictions.

🧠 Prediction on Test Data

Since the test dataset lacks Weekly_Sales, lag and rolling features were generated using the combined train+test data.