CapitalSavvy

View Our Site, Capital Savvy!

Why

We are a group of close friends who are very passionate about machine learning and its application in real problem spaces. As financial transactions become increasingly digitized in the modern age, the attack surface for potential intrusions only increases. Platforms like Google Pay, Apple Pay, and mobile payment apps have provided convenience to consumers at the cost of security. In the field of machine learning, neural networks, a model inspired by the human brain, have seen growing usage due to their versatility and deep comprehension of high-dimensional data. Our group decided to implement an artificial intelligence based on a deep neural network to analyze financial transactions and detect fraud.

What

In the span of 12 hours, we brainstormed, designed, and implemented a full stack web application solving the problem of fraud detection in credit card transactions. We are running a custom Python web server on a virtual private server (VPS) hosted in the cloud. On the backend, we provide access to our deep learning model, which serves as the core of our analysis engine. We implemented our model using TensorFlow, a professional framework for deep learning. We designed an intuitive and simple user interface on our front end with custom styling and graphics in order to make our tool as easy to use as possible.

What we spent a lot of time on

We took great care to validate the accuracy of our model and prevent a phenomenon known as overfitting, where a machine learning model gains high accuracy but does not generalize to all future potential datasets. To combat this, we added dropout layers, used K-Fold validation to ensure unbiased selection of training and testing datasets, and evaluated the problem on a variety of different neural architectures.

"This dataset contains credit card transactions made by European cardholders in the year 2023. It comprises over 550,000 records, and the data has been anonymized to protect the cardholders' identities." -link

Artificial intelligence and technology in general has always had accessibility problems, so it was important for us to make the user interface very intuitive. We spent a lot of time on styling and graphics, for example, making sure that fonts are visible for low-vision users, making our site accessible to all kinds of users.