An end-to-end data analytics project focused on transaction performance, fraud risk and operational insights from 100,000 UPI transactions.
The project analyses customer, merchant, device, transaction and fraud-alert data to identify:
- Transaction success and failure patterns
- Fraud-risk indicators
- High-risk devices and merchants
- Fraud-alert resolution performance
- Regional and channel-level trends
- 100,000 transactions analysed
- 92.14% transaction success rate
- 5.87% transaction failure rate
- 2.00% overall fraud rate
- Rooted-device fraud rate: 20.69%
- Non-rooted-device fraud rate: 1.39%
- 87.60% of fraud alerts were resolved
- 248 fraud alerts remained unresolved
The strongest finding was that rooted devices had nearly 15 times the fraud rate of non-rooted devices.
- Excel – data validation and cleaning
- MySQL – joins, CTEs, views and business queries
- Python – data cleaning, EDA and visualisation
- SciPy – statistical hypothesis testing
- Power BI – executive and fraud dashboards
- HTML, CSS and JavaScript – interactive portfolio website
- Data validation in Excel
- Database design and analysis in SQL
- Exploratory data analysis in Python
- Statistical hypothesis testing
- Dashboard development in Power BI
- Interactive website creation
- Detect and restrict risky activity from rooted devices
- Introduce step-up authentication for high-risk transactions
- Review merchants with unusually high fraud rates
- Investigate unresolved fraud alerts
- Improve ownership of transaction failure reasons
Akshat Raghav