SignaSure

Inspiration

With two members in the group being closely connected to the banking industry, check fraud was on our radar as a problem that needed to be addressed. It costs the financial industry billions annually, so we sought to improve on existing signature verification technologies.

What it does

It reads in a real signature and a signature the user is unsure of, then returns whether it thinks the signature is genuine or fraudulent.

How we built it

We trained a CNN on a kaggle data set of genuine and forged signatures, then integrated it with an API to a react frontend where users can demo the product.

Challenges we ran into

The first model we used employed a siamese neural network, but was ultimately too small architecture-wise for our intended purpose. Another significant setback was the struggle to host the api on AWS EC2, which timed out for seemingly no reason.

Accomplishments that we're proud of

We also were able to train a 92% accurate model. We learned react, tensorflow, and flask despite having no previous experience in any of the three.

What we learned

Learning new web development, Machine learning, and front end.

What's next for SignaSure

Learning how to host the API and create a distributable api for users to employ locally. We would also like to fix the Siamese neural network to improve accuracy.