🛒 Retail Sales Forecasting & Inventory Optimization System

An end-to-end retail analytics system that forecasts product demand using machine learning and optimizes inventory using operations research formulas — deployed as an interactive streamlit simulation dashboard.

---

📌 Problem Statement

Retail businesses lose $1.75 trillion globally due to overstocking and stockouts (IHL Group, 2023). Traditional manual stock planning cannot handle thousands of products across multiple stores. This project builds a data-driven system to predict demand and automate inventory decisions.

💡 Business Value

Problem	This System's Solution
Overstock (excess inventory costs money)	EOQ calculation minimizes order costs
Stockout (lost sales, lost customers)	Reorder point alert ensures timely replenishment
Manual forecasting is slow & inaccurate	XGBoost model achieves <10% MAPE
No visibility into inventory health	Taipy dashboard shows real-time status

🏭 Industry Relevance

Companies like D-Mart, Amazon, Walmart, Flipkart, and Reliance Retail use similar systems for:

• Demand sensing and weekly procurement planning
• Category-level inventory optimization
• Festival/seasonal stock preparation
• Supplier replenishment automation

---

🧰 Tech Stack

Component	Tool
Language	Python 3.10+
Data Processing	Pandas, NumPy
Machine Learning	XGBoost, Scikit-learn
Visualization	Matplotlib, Seaborn, Plotly
Dashboard & Simulation	Taipy (stateful, scenario-ready)
Notebooks	Jupyter
Version Control	Git + GitHub

---

🏗️ Architecture

Raw CSV Data → Preprocessing → Feature Engineering → XGBoost Model
                                                           ↓
                                                   30-Day Forecasts
                                                           ↓
                                           Inventory Optimization Engine
                                           (Safety Stock, Reorder Point, EOQ)
                                                           ↓
                                              streamlit Interactive Dashboard
                                              (Charts + Scenario Simulation)

---

📁 Folder Structure

Retail-Sales-Forecasting-Inventory-Optimization/
├── data/raw/              # Synthetic dataset
├── data/processed/        # Cleaned + featured data
├── notebooks/             # Jupyter EDA and modeling notebooks
├── src/                   # Modular Python source code
├── models/                # Saved ML models (.pkl)
├── outputs/               # Charts, forecasts, recommendations
├── app/                   # streamlit dashboard
├── images/                # Screenshots for README
├── reports/               # Business summary
├── main.py                # Run full pipeline
└── requirements.txt

---

⚙️ Installation

# Clone repository
git clone https://github.com/YOUR_USERNAME/Retail-Sales-Forecasting-Inventory-Optimization.git
cd Retail-Sales-Forecasting-Inventory-Optimization

# Create virtual environment
python -m venv retail_env
source retail_env/bin/activate  # Windows: retail_env\Scripts\activate

# Install dependencies
pip install -r requirements.txt

---

🚀 How to Run

# Run full pipeline (data → model → inventory → charts)
python main.py

# Launch interactive streamlit dashboard
cd app && python dashboard.py
# → Open http://localhost:5000

---

📊 Dataset Details

Feature	Description
Rows	~109,500 daily sales records
Time Period	January 2021 – January 2024 (3 years)
Stores	5 (Mumbai, Delhi, Bangalore, Chennai, Hyderabad)
Products	20 across 5 categories
Synthetic Patterns	Seasonality, weekend effects, festival spikes, stockouts

---

🤖 Model Performance

Model	RMSE	MAE	MAPE
XGBoost	~4.2	~3.1	~8.3%
Random Forest	~5.1	~3.8	~10.2%

MAPE < 15% is considered good for retail demand forecasting.

---

📦 Inventory Optimization Formulas

Safety Stock  = Z × σ_demand × √(lead_time)
Reorder Point = (avg_daily_demand × lead_time) + safety_stock
EOQ           = √(2 × annual_demand × ordering_cost / holding_cost)

---

🖼️ Screenshots

Dashboard

Monthly Sales Trend

Forecast vs Actual

Reorder Recommendations Table

Inventory Health

---

🧪 Simulation Workflow

1. Generate 3-year synthetic retail dataset
2. Clean and preprocess data (handle nulls, outliers, types)
3. Engineer 25+ features (lag, rolling, seasonality, price)
4. Train XGBoost forecasting model (time-series split)
5. Compute safety stock, reorder points, EOQ per product
6. Launch streamlit dashboard and test what-if scenarios

---

🔮 Future Improvements

• Multi-store ensemble forecasting
• Real-time dashboard with live data feed
• Prophet model for better trend+seasonality decomposition
• Promotional impact modeling (price elasticity)
• Weather and event-based demand adjustment
• ERP system integration (SAP/Oracle)
• Anomaly detection for unusual sales spikes

---

🎓 Learning Outcomes

• Time-series feature engineering for ML
• XGBoost regression for demand forecasting
• Operations research inventory formulas (EOQ, ROP, Safety Stock)
• streamlit for building stateful business dashboards
• End-to-end ML pipeline design
• Professional GitHub documentation

---

👤 Author

[CH S K CHAITANYA]

• 📧 chskchaitanya755@gmail.com
• 🔗 LinkedIn
• 🐙 GitHub

Open to Data Analyst, Business Analyst, Retail Analytics, and Data Science roles.