📈 Retail Sales Forecasting

The project demonstrates an end-to-end ML workflow — from raw data ingestion and feature engineering to model evaluation and an interactive Streamlit dashboard.

🚀 Project Overview

Retail sales data is highly volatile and influenced by temporal patterns. This project focuses on predicting daily revenue using historical transaction data and advanced feature engineering, achieving high predictive accuracy.

🔑 Key Highlights

• Leakage-free ML pipeline
• Strong feature engineering (lags, rolling stats, momentum)
• Multiple ML models compared
• Interactive Streamlit UI
• Clean, modular, production-style codebase

🧠 Models Used

Model	Purpose
Random Forest Regressor	Baseline ensemble model
XGBoost Regressor	Final high-performance model

Feature engineering contributed more to performance improvement than adding new models.

📊 Final Model Performance

Model	RMSE	MAE	R²
Random Forest	279.53	175.40	0.909
XGBoost (Best)	240.12	153.10	0.933

✅ XGBoost explains 93.3% of daily revenue variance

🏗️ Project Architecture

retail-sale-forcasting/
│
├── app/
│   └── streamlit_app.py          # Streamlit dashboard
│
├── src/
│   ├── data_loader.py            # Load & aggregate raw data
│   ├── feature_engineering.py    # Lag, rolling, momentum features
│   ├── train_ml_models.py        # Model training
│   ├── evaluate_models.py        # Model evaluation
│   └── metrics.py                # RMSE, MAE, R²
│
├── data/
│   ├── raw/
│   │   └── retail_sales_dataset.csv
│   └── processed/
│       ├── daily_data.csv
│       ├── daily_data_features.csv
│       ├── X_train.csv
│       ├── X_test.csv
│       ├── y_train.csv
│       └── y_test.csv
│
├── models/
│   ├── random_forest.pkl
│   └── xgboost.pkl
│
├── requirements.txt
└── README.md

🔧 Feature Engineering Strategy

To capture real retail behavior, the following features were engineered:

✅ Memory (Lags)

• Revenue lags: 1, 7, 14, 30 days
• Transaction & quantity lags

✅ Trend & Momentum

• Revenue difference (1-day, 7-day)
• Percentage change (7-day)

✅ Volatility

• Rolling standard deviation (7, 14, 30 days)

✅ Smoothed Baselines

• Rolling mean revenue (7, 14, 30 days)

All features use past values only → no data leakage.

🖥️ Streamlit Dashboard

The Streamlit UI provides:

• 📈 Historical daily revenue visualization
• 📊 Model comparison table
• 🏆 Automatic best model selection
• 💼 Business-friendly interpretation of results

▶️ Run the dashboard

streamlit run app/streamlit_app.py

⚙️ Installation & Setup

1️⃣ Clone the repository

git clone <repo-url>
cd retail-sale-forcasting

2️⃣ Install dependencies

pip install -r requirements.txt

3️⃣ Run pipeline (optional)

python src/data_loader.py
python src/feature_engineering.py
python src/train_ml_models.py
python src/evaluate_models.py

💼 Business Interpretation

• The model predicts daily revenue with an average error of ~150 units
• High accuracy achieved through temporal feature engineering
• Suitable for:
  • Sales planning
  • Inventory optimization
  • Revenue trend analysis

🧰 Tech Stack

• Language: Python
• Libraries: Pandas, NumPy, Scikit-learn, XGBoost
• Visualization: Streamlit
• ML Techniques: Feature Engineering, Ensemble Learning, Time-aware Validation