By Dave DeFusco
Now that digital transactions are becoming the norm, from mobile banking to online shopping, the threat of financial fraud is more real than ever. Criminals are constantly finding new ways to steal money, outsmarting traditional fraud detection systems that rely on fixed rules and manual reviews. But what if technology could get smarter, faster and more secure? That’s exactly what three Katz School students in the M.S. in Computer Science set out to explore.
In their Engineering Science & Technology Journal study, “Machine Learning and Blockchain Approaches for Enhancing Fraud Prevention in Financial Transactions,” co-authors Kumbirai Bernard Muhwati, Sarah Mavire and Enock Katenda developed a modern solution to an age-old problem. They combined the predictive power of machine learning with the data security of blockchain technology to build a fraud detection system that’s not only accurate, but nearly impossible to tamper with.
“Financial fraud is constantly evolving. We wanted to build a system that evolves with it,” said Katenda. “Machine learning lets the system learn from patterns, while blockchain keeps everything secure and transparent.”
The team started with a simple observation: traditional fraud detection systems are falling behind. Most banks still rely on fixed rules, like flagging a purchase over a certain amount or coming from a foreign country. But these methods can’t keep up with the fast-changing tactics of cybercriminals. They also generate a lot of false alarms, wasting time and money.
“You don’t want a system that cries wolf every time someone buys a plane ticket,” said Mavire. “But you do want it to stop someone from draining your account at midnight using a stolen login.”
To build their system, the students used a dataset of over 2,500 financial transactions, each labeled as either fraudulent or legitimate. The data included features like: Transaction amount, login attempts, time of transaction, customer occupation and duration of the transaction. They trained two popular machine learning models—Support Vector Machine (SVM) and Random Forest—to learn the patterns that usually show up in fraud cases.
They found that the Random Forest model outshined the SVM, achieving nearly perfect results for accuracy, precision, recall and F1-Score, which is a measure used in machine learning to evaluate how well a classification model performs especially when dealing with imbalanced data, like fraud detection, where fraudulent transactions are rare compared to legitimate ones.
“It was exciting to see the Random Forest model perform so well,” said Muhwati. “It means the system isn’t just accurate in theory, it works in practice, too.”
As important, they discovered that transaction amount, transaction duration and customer occupation were the strongest predictors of fraud. For example, very large or unusually fast transactions often signaled suspicious behavior, but high accuracy wasn’t enough. The students wanted to ensure that fraud detection systems are secure and trustworthy, especially as transactions move between institutions.
That’s where blockchain came in. A blockchain is like a digital notebook that records every transaction. Once something is written, it can’t be changed without everyone noticing. This makes it incredibly difficult for hackers or insiders to alter data without leaving a trace.
“Blockchain adds a layer of trust,” said Mavire. “It ensures that no one can manipulate the data, not even someone inside the system.”
Even better, blockchain allows for real-time auditing, meaning institutions can immediately trace suspicious activity and act faster. By combining machine learning’s ability to predict fraud with blockchain’s ability to secure and verify the data, the study presents a powerful new model for banks and other financial institutions.
“We’re talking about a system that not only catches fraud faster, but also makes sure the data it uses is clean and cannot be tampered with,” said Muhwati. “That’s a game-changer.”
The implications of the Katz School students’ research go beyond academic theory. Banks, fintech companies and even government agencies are all looking for ways to stay ahead of cybercriminals. Practical recommendations from the study include:
- Monitor unusual transaction amounts closely: Since transaction size was a key fraud indicator, systems should automatically flag suspicious spikes.
- Track user behavior patterns: Fast or repetitive login attempts and sudden changes in transaction times can signal fraud.
- Use models that handle imbalanced data: Fraud is rare but costly. Systems must be trained to spot the subtle patterns without raising too many false alarms.
“Our goal was not just to build a tool, it was to offer a roadmap that real companies can follow to better protect their customers,” said Katenda. “Imagine a world where financial institutions work together, share verified data securely and catch fraud before it happens. That’s the vision we’re working toward.”
