Vendor Invoice Intelligence System

AI-Powered Freight Cost Prediction & Invoice Risk Detection

---

📌 Overview

This system helps finance teams automatically:

1. Predict freight costs – Estimate fair shipping costs for vendor invoices
2. Flag risky invoices – Detect anomalies and suspicious patterns

Built with Machine Learning and deployed on Streamlit.

---

🎯 Business Problem

Finance teams spend hours manually reviewing vendor invoices to detect:

• Freight overcharging – Abnormally high shipping costs
• Invoice anomalies – Mismatches between invoice and purchase orders
• Fraudulent patterns – Suspicious quantity or pricing discrepancies

This system automates that process, enabling faster approvals and reducing financial leakage.

---

🔑 Key Features

• 🧠 Regression Model – Predicts freight costs with 97% accuracy
• 🚨 Classification Model – Identifies risky invoices with 89% accuracy
• 🖥️ Interactive Web App – User-friendly interface for instant predictions
• 📊 Confidence Scores – Model confidence level for each prediction
• 🚀 Deployed & Live – Ready to use on Streamlit Cloud

---

🚀 Live Demo

Try it here: https://mayuresh0711-vendor-invoice-intelligence-system-app-sszyjb.streamlit.app/

Screenshots

Homepage

Freight Cost Prediction

Invoice Risk Flagging

---

💡 How It Works

Model 1: Freight Cost Prediction (Regression)

Problem: How do we know if a freight charge is fair?

• Input: Invoice quantity & total amount
• Output: Expected freight cost based on historical patterns
• Business Impact: If actual freight is 30%+ higher than predicted, it may indicate overcharging
• Example: Invoice of $18,500 → Predicted freight: $1,120 → If billed $1,500+, it's flagged

Model 2: Invoice Risk Flagging (Classification)

Problem: Which invoices need manual review before payment?

• Input: Invoice quantity, amount, freight, total purchase details
• Output: "Safe to Auto-Approve" OR "Requires Manual Review" with confidence score
• Business Impact: Prioritizes high-risk invoices for auditor attention, reducing fraud/errors
• Detects: Amount mismatches, unusual freight ratios, suspicious patterns
• Example: Invoice $2,468 vs Purchase Total $2,476 → Large variance → Flagged

---

📊 Data & Training

Data Source: SQLite database with vendor invoices and purchase order data

Features Used:

• Invoice quantity, amount, freight cost
• Purchase order quantities and totals
• Receiving performance metrics

Models:

• Freight prediction: Random Forest Regressor (tested Linear Regression & Decision Tree)
• Risk flagging: Random Forest Classifier with feature scaling

---

📊 Model Performance

Model	Metric	Value
Freight Prediction	MAE	24.11
	R² Score	96.99%
Risk Flagging	Accuracy	89%
	F1-Score	87%

---

🏗️ Project Workflow

1. DATA LOADING → SQLite database
   ↓
2. PREPROCESSING → Handle missing values, feature engineering
   ↓
3. MODEL TRAINING → Regression & Classification models
   ↓
4. EVALUATION → Test performance metrics
   ↓
5. SERIALIZATION → Save models as .pkl files
   ↓
6. INFERENCE API → predict_freight.py & predict_invoice_flag.py
   ↓
7. STREAMLIT APP → Interactive user interface
   ↓
8. LIVE DEPLOYMENT → Streamlit Cloud

---

📁 Project Structure

├── Data/
│   └── inventory.db                    # Source data (SQLite)
├── freight_cost_prediction/
│   ├── data_preprocessing.py
│   ├── modeling_evaluation.py
│   ├── train.py
│   └── models/predict_freight_model.pkl
├── invoice_flagging/
│   ├── data_preprocessing.py
│   ├── modeling_evaluation.py
│   ├── train.py
│   └── models/predict_flag_invoice.pkl
├── inference/
│   ├── predict_freight.py
│   └── predict_invoice_flag.py
├── notebooks/                          # Jupyter exploration notebooks
├── app.py                              # Streamlit application
└── requirements.txt

---

⚡ Quick Start

1. Clone Repository

git clone https://github.com/mayuresh0711/vendor-invoice-intelligence-system.git
cd vendor-invoice-intelligence-system

2. Setup Environment

python -m venv venv

# Windows:
venv\Scripts\activate

# macOS/Linux:
source venv/bin/activate

3. Install Dependencies

pip install -r requirements.txt

4. Train Models (Optional)

python freight_cost_prediction/train.py
python invoice_flagging/train.py

5. Run Application

streamlit run app.py

Open: http://localhost:8501

---

🛠️ Technologies Used

• Python – Core language
• scikit-learn – Machine learning
• Pandas & NumPy – Data manipulation
• Streamlit – Web application
• SQLite – Database

---

📞 Contact

Mayuresh Ahire
Data Analyst | Machine Learning

• 🔗 LinkedIn: linkedin.com/in/mayuresh-ahire-ab079b2a3/
• 🐙 GitHub: github.com/mayuresh0711
• 📧 Email: ahiremayuresh4@gmail.com

---

Happy to connect! Feel free to reach out with feedback or collaboration ideas. 🚀