Vendor Invoice Intelligence System

Freight Cost Prediction & Invoice Risk Flagging

📌 Table of Contents

• Project Overview
• Business Objectives
• Data Sources
• Exploratory Data Analysis
• Models Used
• Evaluation Metrics
• Application
• Project Structure
• How to Run This Project
• Author & Contact

---

📌 Project Overview

This project implements an end-to-end machine learning system designed to support finance teams by:

1. Predicting expected freight cost for vendor invoices.
2. Flagging high-risk invoices that require manual review due to abnormal cost, freight, or operational patterns.

---

🎯 Business Objectives

1. Freight Cost Prediction (Regression)

Objective:
Predict the expected freight cost for a vendor invoice using quantity, invoice value, and historical behavior.

Why it matters:

• Freight is a non-trivial component of landed cost.
• Poor freight estimation impacts margin analysis and budgeting.
• Early prediction improves procurement planning and vendor negotiation.

2. Invoice Risk Flagging (Classification)

Objective:
Predict whether a vendor invoice should be flagged for manual approval due to abnormal cost, freight, or delivery patterns.

Why it matters:

• Manual invoice review does not scale.
• Financial leakage often occurs in large or complex invoices.
• Early risk detection improves audit efficiency and operational control.

📂 Data Sources

Data is stored in a relational SQLite database (inventory.db) with the following tables:

• vendor_invoice – Invoice-level financial and timing data
• purchases – Item-level purchase details
• purchase_prices – Reference purchase prices
• begin_inventory, end_inventory – Inventory snapshots

SQL aggregation is used to generate invoice-level features.

---

📊 Exploratory Data Analysis (EDA)

EDA focuses on business-driven questions, such as:

• Do flagged invoices have higher financial exposure?
• Does freight scale linearly with quantity?
• Does freight cost depend on quantity?

Statistical tests (t-tests) are used to confirm that flagged invoices differ meaningfully from normal invoices.

---

🤖 Models Used

Regression (Freight Prediction)

• Linear Regression (baseline)
• Decision Tree Regressor
• Random Forest Regressor (final model)

Classification (Invoice Flagging)

• Logistic Regression (baseline)
• Decision Tree Classifier
• Random Forest Classifier (final model with GridSearchCV)

Hyperparameter tuning is performed using GridSearchCV with F1-score to handle class imbalance.

---

📈 Evaluation Metrics

Freight Prediction

• MAE
• RMSE
• R² Score

Invoice Flagging

• Accuracy
• Precision, Recall, F1-score
• Classification report
• Feature importance analysis

---

🖥 End-to-End Application

A Streamlit application demonstrates the complete pipeline:

• Input invoice details
• Predict expected freight
• Flag invoices in real time
• Provide human-readable explanations

---

📁 Project Structure

inventory-invoice-analytics/
│
├── data/
│   └── inventory.db
│
├── freight_cost_prediction/
│   ├── data_preprocessing.py
│   ├── model_evaluation.py
│   └── train.py
│
├── invoice_flagging/
│   ├── data_preprocessing.py
│   ├── model_evaluation.py
│   ├── model_evaluation.py
│   └── train.py
│
├── inference/
│   ├── predict_freight.py
│   └── predict_invoice_flag.py
│
├── models/
│   ├── predict_freight_model.pkl
│   ├── scaler.pkl
│   └── predict_flag_invoice.pkl
│
├── notebooks/
│   ├── Invoice Flagging.pkl
│   └── Predict Freight Cost.ipynb
│
├── app.py
├── README.md
└── .gitignore

---

How to Run This Project

1. Clone the repository:

git clone https://github.com/yourusername/inventory-invoice-analytics.git

2. Train and Save Best Fit Models:

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

3. Test Models:

python inference/predict_freight.py
python inference/predict_invoice_flag.py

4. Open Application:

streamlit run app.py

---

Author & Contact

Nimesh Patel
ML Engineer
📧 Email: pnimesh0806@gmail.com