An end-to-end data analytics and fraud risk intelligence project built using Python, SQL, MySQL, Power BI, and DAX.
The project analyzes UPI transaction data to identify fraud patterns, understand transaction risk drivers, and present actionable insights through an interactive Power BI dashboard.
UPI transactions generate large volumes of financial data, making it important to identify unusual transaction patterns and potential fraud.
This project follows an end-to-end analytics workflow:
Raw Data → Data Cleaning → Feature Engineering → SQL Analysis → Risk Scoring → Power BI Dashboard → Business Insights
The objective is to analyze transaction behavior and create a practical risk intelligence system rather than simply performing descriptive analysis.
| Metric | Value |
|---|---|
| Total Transactions | 250,000 |
| Fraud Transactions | 480 |
| Fraud Rate | 0.192% |
| High-Risk Transactions | ~11,000 |
| Data Period | January–December 2024 |
The dataset contains a highly imbalanced fraud class, making fraud-rate analysis and risk-based segmentation particularly important.
- Python — Data cleaning, preprocessing, feature engineering and analysis
- Pandas & NumPy — Data manipulation and numerical analysis
- SQL — Risk analysis and transaction-level investigation
- MySQL — Database storage and querying
- Power BI — Interactive dashboard and visualization
- DAX — Measures, KPIs and risk calculations
- Jupyter Notebook — Exploratory data analysis
- Git & GitHub — Version control and project documentation
UPI_Transaction_Risk_Intelligence/
│
├── data/
│ ├── raw/
│ └── processed/
│
├── notebooks/
│ └── 01_data_understanding.ipynb
│
├── reports/
│ └── UPI_Transaction_Risk_Intelligence.pbix
│
├── sql/
│ └── 01_risk_analysis
│
├── src/
│ └── load_to_mysql.py
│
├── .gitignore
├── README.md
└── requirements.txt
Note: The transaction CSV files are intentionally excluded from this GitHub repository because of their size. They are kept locally and are covered by
.gitignore.
The project uses a synthetic UPI transaction dataset containing approximately 250,000 transactions across 2024.
The dataset includes transaction and behavioral attributes such as:
- Transaction ID
- Timestamp
- Transaction type
- Merchant category
- Transaction amount
- Transaction status
- Sender age group
- Receiver age group
- Sender state
- Sender bank
- Receiver bank
- Device type
- Network type
- Fraud flag
The dataset contains 480 fraud transactions, representing approximately 0.192% of all transactions.
The data preparation workflow included:
- Loading the raw transaction data using Python.
- Checking dataset dimensions and column information.
- Converting timestamp fields into appropriate datetime format.
- Checking for missing values.
- Checking categorical values for inconsistencies.
- Validating transaction statuses and categories.
- Performing exploratory data analysis.
- Creating additional features for risk analysis.
- Generating transaction risk scores and risk tiers.
The project investigates fraud and transaction risk across multiple dimensions, including:
- Transaction type
- Merchant category
- Device type
- Network type
- Sender state
- Transaction status
- Transaction amount
- Time of transaction
- Risk tier
Because fraud represents a very small proportion of total transactions, the analysis focuses on both fraud patterns and broader risk indicators.
MySQL is used to perform structured transaction analysis and investigate risk patterns.
The SQL analysis includes:
- Transaction volume analysis
- Fraud transaction analysis
- Fraud-rate calculations
- Category-level comparisons
- Risk segmentation
- Transaction investigation queries
The SQL scripts are available in the sql/ directory.
The project includes an interactive Power BI dashboard consisting of three main pages.
Provides a high-level view of:
- Total transactions
- Fraud transactions
- Fraud rate
- High-risk transactions
- Overall transaction risk
Analyzes factors associated with transaction risk, including:
- Transaction type
- Merchant category
- Device type
- Network type
- Sender state
- Risk tier
Provides a more detailed transaction-level view for investigating individual transactions and their associated risk characteristics.
Power BI File
The complete Power BI dashboard is available here:
UPI_Transaction_Risk_Intelligence.pbix
Some observations from the analysis include:
- The dataset contains a highly imbalanced fraud class, with fraud accounting for approximately 0.192% of transactions.
- Most transactions are successful, while failed transactions form a smaller proportion of the dataset.
- Fraud patterns vary across transaction types and merchant categories.
- Risk segmentation helps identify a much broader set of potentially risky transactions than confirmed fraud cases alone.
- Transaction-level analysis can be used to investigate individual risk indicators.
A risk-scoring approach was developed to categorize transactions into different risk levels.
Example risk tiers:
- Low Risk
- Medium Risk
- High Risk
The risk tier is intended as an analytical risk indicator and should not automatically be interpreted as confirmed fraud.
The Python workflow includes:
Raw Dataset
↓
Data Understanding
↓
Data Cleaning
↓
Feature Engineering
↓
Risk Scoring
↓
Processed Dataset
The main exploratory notebook is available in:
notebooks/01_data_understanding.ipynb
The Python loader connects the processed transaction data to MySQL.
The database workflow is:
Processed Data
↓
Python MySQL Loader
↓
MySQL Database
↓
SQL Risk Analysis
Database credentials are stored locally using environment variables and are not included in this repository.
git clone https://github.com/ria-space/UPI-Transaction-Risk-Intelligence.git
cd UPI-Transaction-Risk-Intelligencepython -m venv .venvWindows:
.venv\Scripts\activatepip install -r requirements.txtCreate a local .env file containing:
MYSQL_HOST=localhost
MYSQL_USER=YOUR_MYSQL_USERNAME
MYSQL_PASSWORD=YOUR_MYSQL_PASSWORD
MYSQL_DATABASE=upi_risk_intelligence
Do not upload the .env file to GitHub.
The following files are intentionally excluded from GitHub:
- Raw transaction CSV files
- Processed transaction CSV files
.env- Python virtual environment
This keeps the repository lightweight and prevents credentials from being exposed.
The main objective of this project is to demonstrate an end-to-end data analytics and fraud risk intelligence workflow using real-world-style transaction data.
It combines:
Python + SQL + MySQL + Power BI + DAX
to transform raw transaction data into analytical insights and an interactive risk dashboard.
Data Science Student
GitHub: @ria-space