About the Project

Inspiration

UPI has made digital payments in India extremely convenient, but it has also increased fraud. Many users only realize something is wrong after losing money. I wanted to build a solution that helps users identify fraud before approving a payment, not after.

What it does

FraudX analyzes UPI and online payment requests in real time and assigns a Risk Score:

• Safe
  
• Caution
  
• High Risk
  

It detects unusual patterns such as:

• New payees
  
• Odd timings
  
• Abnormal frequency
  

FraudX also explains why a transaction may be risky, helping users make informed decisions.

How I built it

I built an MVP using Python, simulating common fraud scenarios.

• A lightweight detection engine identifies anomalies.
  
• A simple frontend displays risk scores and clear alerts for users.
  

The anomaly detection logic is based on statistical thresholds. For example, if transaction frequency exceeds a baseline mean (\mu) by more than two standard deviations (\sigma), i.e.:

[ \text{Risk Trigger} = \frac{X - \mu}{\sigma} > 2 ]

then the transaction is flagged as suspicious.

Challenges I faced

• Lack of access to real-world fraud data, requiring realistic simulations
  
• Balancing detection accuracy with real-time performance
  
• Designing alerts that warn users without overwhelming them
  

Accomplishments

• Built a working real-time fraud detection MVP
  
• Designed an intuitive, easy-to-understand interface
  
• Created a scalable foundation that can extend beyond UPI
  

What I learned

Fraud prevention isn’t just a technical problem—clarity, trust, and user experience are just as important. Even lightweight models can be effective when designed thoughtfully.

What’s next for FraudX

• Improve accuracy with richer datasets
  
• Expand platform support (mobile + browser)
  
• Explore partnerships with banks and UPI providers to bring proactive fraud detection to more users

Built With

• browser
• extension
• python
• scikit-learn
• tensorflow-lite