CreditShield

CreditShield is a tool that monitors bank transactions for fraud in real time. Technologies like machine learning and synthetic data generation are integrated in order to be as accurate as possible.

Comment 1

Inspiration

BNY's challenge was immediately interesting to us, as technology like this is becoming increasingly important. The more that we can integrate things like AI into services that help and protect people, the better off we'll all be.

What it does

The core of our project is the machine learning model itself. After training on a set of synthetic data that we created, it can then examine new transactions in an existing bank account and determine how likely they are to be fraudulent. Along with the model, we also worked on a real time data stream using Apache Kafka. Our solution for opening ledgers utilized blockchain and Ethereum for data integrity, decentralization, and scalability.

How we built it

We broke up different components into subprojects and each person worked on one. These were the machine learning model, synthetic data generation, blockchain integrity, and the real time data stream.

Challenges we ran into

Changing parts of the project and not having figured out everything before starting were the main issues that we ran into. We were communicating as a team and spent lots of time together, but for the first day a lot of what we did was individual work and not all of it came together seamlessly.

Accomplishments that we're proud of

Making sure the model was accurate, and had a strong sense of pattern recognition through the engineering of features that weren't first presented in our dataset. These engineered features include the typical days of the week a person initiates their transaction, their usual times, Z-score between the amount they spend during those typical times, and last, if the company of their transaction actually exists. Creating these relational features and making sure the model training doesn't over-fit to the training set was extremely rewarding and contributed to more accurate prediction!

Our other accomplishments include having a good looking user interface, and a complex system for generating the synthetic data.

What we learned

Some of the main lessons we took away were that having a clear plan and knowing what the final product should look like are both incredibly important for fast paced projects like this.

As far as new programming skills, we got more experience with machine learning, our first exposure to synthetic data generation, OpenAI API, Apache Kafka, and blockchain.

What's next for CreditShield

We're happy with how our project came out in the end and for now don't have more to add, but we'll certainly all be revisiting ideas that we learned more about while working on it.

Built With

Share this project:

Updates