Inspiration

There In today’s digital world, the issue of cybersecurity goes beyond an issue for enterprises and becomes an issue for each individual as well. None of us have not noticed the headlines in the last year—phishing attacks are up 150% and cost users billons of dollars.

The motivation for creating PhishGuard began with a personal need—protection for non-technical relatives, such as elderly relatives who have difficulty in differentiating between an authentic message from the bank and an attack from scammers. We wanted to create a service that would be a “second pair of eyes” with the help of Artificial Intelligence.

What it does

PhishGuard is a real-time online application to analyze URLs in order to identify possible phishing attacks.

Instant Analysis - Users provide a URL, which is instantly evaluated.

AI-Assisted Detection: Unlike the classical anti-phishing tools that only scan the "blacklist," PhishGuard employs Machine Learning Technology to examine the actual URL. It can identify Zero Day attacks (emerging scams) that are not yet known.

Transparency: It’s not simply “Bad,” it gives the confidence level and the reasons behind the link being bad.

How we built it

We adopted a "Low-Code, High-Impact" approach, which concentrated on the core of the machine learning process.

Language: Python 3.

Interface: - Created using Streamlit.

Machine Learning Algorithm We used the Scikit-Learn library. Our solution consists of a Logistic Regression Engine coupled with a TF-IDF Vectorizer.

Issues encountered

The team encountered the following challenges

Data Imbalance - Finding a data set with a balanced mix of safe and malicious URLs proved to be tricky. Real-world datasets tend to be skewed.

Feature Extraction: "URLs are just text strings," according to the book. We needed to learn how to appropriately tokenize these strings to handle instances such as "secure-login" in the wrong location.

Accomplishments that we’re proud of

Completing the implementation of an NLP pipeline end-to-end.

Developing an user interface which is simple enough for a layman to use but is based on complicated mathematics.

Reaching a high accuracy level with our validation set in testing.

What We Learned - we learned on how to integrate different fields of knowledge to create something useful ## What’s next for PhishGuard - Real-Time AI Threat Detector Browser Extension:

Our proposed plan is to convert this model into a chrome extension to auto-scan the links on hover. API Integration: We will make our scanning engine available for use by other API developers. Deep Learning: Testing LSTM (Long Short-Term Memory) networks to enhance detection of even more obfuscated URLs.