🛒 Retail Demand Forecasting using Machine Learning

A machine learning project for forecasting weekly retail demand across multiple supermarkets using historical sales and promotional data.

---

📌 Project Overview

Accurate demand forecasting is essential for effective inventory management in retail businesses. Poor demand estimation can lead to:

• 📉 Stock-outs (lost sales)
• 📦 Excess inventory
• 💰 Increased storage costs
• 📊 Inefficient supply chain planning

This project develops a machine learning-based forecasting pipeline to predict weekly product demand for different supermarket-SKU combinations 8 weeks in advance.

The complete workflow includes:

• Data Cleaning
• Exploratory Data Analysis (EDA)
• Feature Engineering
• Walk-Forward Time-Series Validation
• LightGBM Forecasting Model
• Business Performance Evaluation

---

🎯 Problem Statement

The objective is to forecast weekly demand for every supermarket and product (SKU) combination using historical demand and promotional information.

The model is designed to:

• Forecast demand 8 weeks ahead
• Capture seasonality and demand trends
• Measure promotion impact
• Support inventory planning decisions

---

📂 Repository Structure

Retail-Demand-Forecasting/

│── Demand_Forecasting.ipynb
│── README.md
│── requirements.txt

├── data/
│   ├── demand.csv
│   └── promotions.csv

├── outputs/
│   ├── prediction_results.csv
│   ├── feature_importance.png
│   └── forecast_example.png

└── models/
    └── final_model.pkl

---

📊 Dataset

The project uses two datasets.

demand.csv

Contains historical weekly demand information.

Features include:

• Date
• Supermarket
• SKU
• Weekly Demand

---

promotions.csv

Contains promotional campaign information.

Features include:

• Promotion Start Date
• Promotion Duration
• Supermarket
• SKU

---

⚙️ Project Workflow

1️⃣ Data Cleaning

• Missing value treatment
• Duplicate checking
• Outlier detection
• Data consistency verification

---

2️⃣ Exploratory Data Analysis

Performed detailed analysis to understand:

• Demand distribution
• Weekly trends
• Seasonality
• Promotion impact
• SKU-wise demand
• Supermarket-wise demand

---

3️⃣ Feature Engineering

Created forecasting features including:

• Lag Features
• Rolling Statistics
• Promotion Indicators
• Calendar Features
• Historical Demand Patterns

---

4️⃣ Model Development

The forecasting model was built using LightGBM Regressor.

Reasons for selecting LightGBM:

• Fast training
• Handles non-linear relationships
• Works well on tabular datasets
• Robust for forecasting problems

---

5️⃣ Time-Series Validation

Instead of using a random train-test split, this project implements Walk-Forward Validation, which better represents real-world forecasting scenarios.

Benefits include:

• No future information leakage
• Realistic evaluation
• Better estimation of production performance

---

📈 Model Evaluation

The model was evaluated using multiple forecasting metrics.

Metric	Value
MAE	11.99
RMSE	18.49
R² Score	0.9787
MAPE	2.21%
WAPE	2.27%

A simple Naive Forecast was also implemented as a baseline for comparison.

---

📉 Feature Importance

The model identifies the most influential variables affecting demand.

Key features include:

• Historical Demand
• Promotion Activity
• Calendar Features
• Lag Variables
• Rolling Statistics

Feature importance visualization is included in the notebook.

---

💼 Business Insights

The analysis provides several practical insights:

• Promotional campaigns increase product demand.
• Historical demand is the strongest predictor.
• Weekly seasonality significantly affects sales.
• Demand patterns vary across supermarkets and products.

These insights can help businesses improve:

• Inventory Planning
• Stock Replenishment
• Warehouse Management
• Promotion Strategy

---

🛠️ Technologies Used

• Python
• Pandas
• NumPy
• Matplotlib
• Scikit-Learn
• LightGBM
• Joblib

---

🚀 How to Run

Clone Repository

git clone https://github.com/Raghav1507/Retail-Demand-Forecasting.git

---

Install Dependencies

pip install -r requirements.txt

---

Launch Jupyter Notebook

jupyter notebook

Open:

Demand_Forecasting.ipynb

---

📌 Future Improvements

Potential enhancements include:

• Hyperparameter Optimization
• Holiday and Festival Features
• Weather Data Integration
• Ensemble Models
• Deep Learning (LSTM/Transformer)
• Automated Forecast Dashboard

---

👨‍💻 Author

Raghav Singhal

B.Tech CSE (Artificial Intelligence & Machine Learning)

Interested in:

• Machine Learning
• Data Science
• Time-Series Forecasting
• Predictive Analytics

GitHub: https://github.com/Raghav1507

---

⭐ Acknowledgements

This project was developed as part of a demand forecasting case study to demonstrate practical machine learning techniques for retail inventory management.