noFraud

Improving our cybersecurity one step at a time.

Comment 1

Inspiration

To find what our team would solve, we began conducting research to determine prevalent problems in our community and the world. we soon found appealing statistics on phishing, fraud, and scam emails. Over 3 billion people are victim to fraud, phishing, and scam emails every year. In Loudoun alone, hundreds of families lose things such as money and valuable items. All our team members have also personally been affected by malicious emails, from scam to fraud. While competing, a team member even received a suspicious text with a fraudulent message and link. For these reasons, we are extremely passionate about this problem.

What it does

Our solution is noFraud, a Machine-learning driven, web-based application hosted on Amazon Web Services to:

  1. classify spam, fraud, and phishing emails using machine learning models
  2. educate the public about fraud, spam, and phishing through online quizzes and informational web pages.

How we built it

Using Python, Flask, Tensorflow, Keras, and the BERT algorithm (advanced NLP model for understanding text by Google), the advanced neural networks were built. After training for roughly an hour, the model could classify spam, phishing, and fraudulent emails with a 99.07% accuracy. Flask was used to build a custom back-end API to host the model for prediction, allowing anyone to classify their spam, fraud, or phishing emails using url-parameters, free of charge.

Using Amazon Web Services, HTML, CSS, and Javascript, the team built a secure website to host the prediction model, information about phishing, spam, and fraud emails, and a quiz to test users on their knowledge about these malicious attacks.

Challenges we ran into

Initially, we began training our machine learning model locally, which had an estimated training time of over 24 hours, since we were using the BERT algorithm to allow the computer to understand the emails. After researching solutions, we were able to train the model on Google Cloud in roughly 1 hour.

Accomplishments that we're proud of

We are proud of our high achieving machine learning model (99.07% accuracy) and the website we are hosting on Amazon Web Services. Our secure website allows anybody to learn about becoming more secure, and our model allows community members to enhance their security with regards to emails.

What we learned

We learned a great deal about spam, phishing, and fraud emails, and hope to apply our knowledge in these fields in the future.

What's next for noFraud

Currently, the back end and front end of noFraud are not fully integrated together. Using React, we can add the back-end API for prediction directly to the noFraud website, allowing for users to classify their emails as phishing, fraud, spam, or benign immediately.

Share this project:

Updates