TransactSecure

TransactSecure_overview

Inspiration

J.P. Morgan’s interbank Quorum blockchain used in other nations across the globe.

What it does

Our solution provides a machine learning-based credit rating system that flags suspicious recipients to the sender and alerts a moderator to take further action if required. Additionally, it leverages blockchain technology to share data across multiple bank institutions, significantly reducing the chances of day-to-day transfer fraud.

How we built it

To create a robust scam detection model, we generated a synthetic dataset of 1,000 transaction records. Each record included relevant transaction details, such as credit rating and transaction amount. We built a data pipeline that preprocesses this data for machine learning, enabling the model to classify transactions as low, medium, or high risk.

For the frontend, we used Svelte and SvelteKit, paired with TailwindCSS for styling the website. On the backend, we utilized PocketBase as an all-in-one service, functioning as both the DAO layer and database layer, which allowed for quick prototyping.

Challenges we ran into

Generating synthetic data posed a significant challenge, requiring careful consideration of feature selection for our machine learning models. Our team members brought diverse strengths and varying familiarity with different frameworks to the project. Initially, we chose Svelte for the project due to its familiarity with our primary web developer. However, when he fell ill, other team members, less experienced with Svelte, had to step in and take over his responsibilities. This unexpected shift necessitated a rapid learning curve, which slowed our progress. Despite these obstacles, we persevered and successfully launched the prototype. This experience underscored the importance of adaptability, teamwork, and the willingness to learn new technologies under pressure.

Accomplishments that we're proud of

Our machine learning model (random forest) achieved an impressive accuracy of 99.5% on the testing data, demonstrating its effectiveness in detecting fraudulent transactions. Despite having other responsibilities on top of this competition, we managed to push the prototype out within the time constraints. This accomplishment highlights our team's dedication and ability to deliver high-quality results under pressure.

What we learned

Effective feature engineering and building high-performing models require iterative development and continuous tuning. Despite challenges faced and our other responsibilities, we successfully pushed out the prototype, highlighting the importance of teamwork and the ability to quickly learn new technologies under pressure.