Phishing Classifier: Detecting Fraudulent Websites Using AI

In today's tech-driven world, cyberattacks are an ever-present threat. One of the most common and successful forms of attack is phishing. Phishing is a type of fraud where an attacker impersonates a legitimate entity—usually through emails, text messages, or social media—aiming to steal sensitive information like personal details or credit card information. Typically, these attacks involve redirecting victims to fake websites designed to look authentic, tricking them into providing personal data or paying for services they’ll never receive. Phishing can also serve as a vehicle for more malicious attacks, such as ransomware.

The high success rate of phishing is often due to the simplicity of its execution, the availability of tools that facilitate website cloning, and the lack of cybersecurity awareness, particularly among vulnerable groups like the elderly and young children. For these reasons, the development of automatic detection systems for fraudulent websites has become crucial to help protect users.

This project focuses on applying and improving artificial intelligence techniques to detect phishing websites, aiming to enhance the accuracy and performance of current models. Various machine learning (ML) algorithms, such as Logistic Regression and Random Forest, along with deep learning techniques like Convolutional Neural Networks (CNNs), will be explored.

Repository Overview

This repository contains the code for six experiments exploring both traditional ML and deep learning models for phishing website classification:

1. Execution of Classical Machine Learning Models:

   • A baseline experiment using classical models (Decision Tree, Random Forest, Logistic Regression, Naïve Bayes, and Support Vector Machine).
   • Script: classical_ml_models.py

2. Dataset Scaling and Repetition of Experiment 1:

   • Standardizing the dataset and re-evaluating the models.
   • Script: scaled_data_models.py

3. Voting Ensemble of Random Forest, Logistic Regression, and Support Vector Machine:

   • Combining the outputs of these models to enhance accuracy.
   • Script: voting_ensemble.py

4. Neural Network with a Single Hidden Layer:

   • Implementing a simple feedforward neural network.
   • Script: simple_nn.py

5. Convolutional Neural Network (CNN):

   • Implementing a CNN model for image-based feature extraction and classification.
   • Script: cnn_model.py

Getting Started

Prerequisites

To run the experiments, ensure you have the following installed:

• Python 3.1x
• Required libraries (see requirements.txt for details)

Running the Experiments

Each experiment is contained in its own script. To run an experiment, execute the relevant script. For example, to run the CNN experiment:

python3 cnn_model.py

Additional Content

Image Preparation

The images used in the CNN model are generated by the generate_images.py script and stored in the figures/ folder. You can regenerate these images by running the generate_images.py script.

python3 generate_images.py

Pre-generated images from the generate_images.py script are stored as follows:

• Training set:

  • Legitimate: figures/train/legitimate/ (4585 images)
  • Phishing: figures/train/phishing/ (4559 images)

• Test set:

  • Legitimate: figures/test/legitimate/ (1130 images)
  • Phishing: figures/test/phishing/ (1156 images)

These images are used as input to the CNN model in cnn_model.py for training and testing purposes.

Explainable AI (XAI)

To enhance transparency and trust in the model predictions, we use XAI methods to identify the most significant features contributing to the classification of websites. This analysis provides insights into how the model differentiates between legitimate and phishing websites. You can run the XAI analysis with the following command:

python3 xai_analysis.py

Model Summary

Experiment	Model Type	Key Techniques	Script
Classical ML	Decision Tree (DT), Random Forest (RF), Logistic Regression (LR), Naïve Bayes (NB), Support Vector Machine (SVM)	Baseline performance	classical_ml_models.py
Scaled Data	DT, RF, LR, NB, SVM	Data standardization	scaled_data_models.py
Voting Ensemble	RF, LR, SVM	Model ensemble for enhanced accuracy	voting_ensemble.py
Simple Neural Network	Single Hidden Layer	Basic neural network structure	simple_nn.py
Convolutional Neural Network	Convolutional Neural Network (CNN)	Feature extraction for images	cnn_model.py