Phishing URL Detection Using Ensemble Learning

Abstract

The conflict of having to become safer on the internet is one. continually expanding one, and phishing attacks are a huge part of concern. As for the current work, several aspects of supervised learning are explored in the present paper. Decision Tree and others are generally very handy when it comes to detecting phishing links. RF has been developed in computer science along with Support Vector Machine (SVM), Naive Bayes, k-nearest Neighbors (KNN), and linear support vector classifier. (LSVC). The validity of each model is then discussed and as well their strength and possible weaknesses as in identifying These factors are used especially, targets of phishing attempts are highlighted. So to improve the results further, we recommend an ensemble of three classifiers – Random Forest, Logistic Regression, and Gradient Boost, the top three models. It is an ensemble technique that uses predictions of several base estimators to enhance the sorte of generality or stability compared to another estima to or, for a model compared to the others or compared to a multivariat The study also reveals that ensemble design strategies enhance the performance of the systems. and as such enhance the current phishing detection systems with better defenses against such attacks. Therefore, the results of this research will improve the advance. measures related to security and greatly stress the advisability of using various methods existing and in use in the contemporary currents of cyber security

Datasets

1. Phishing and Legitimate URLs Dataset

• Source: Kaggle
• Description: This dataset contains a collection of phishing and legitimate URLs, labeled for supervised learning tasks.
• Features:
  • 88,647 phishing URLs
  • 48,932 legitimate URLs
  • 9 Features including URL length, domain registration, and more.

2. PhishDataset

• Source: GitHub
• Description: Another comprehensive dataset consisting of phishing and legitimate URLs with a diverse range of features.
• Features:
  • 27,000 phishing URLs
  • 35,000 legitimate URLs
  • Features such as domain age, the presence of suspicious characters, and domain-based attributes.

Project Objectives

• Preprocess the data: Clean and preprocess the datasets to remove null values and perform feature engineering.
• Feature extraction: Extract relevant features from the URLs such as length, number of dots, special characters, and domain age.
• Modeling: Implement various machine learning models like Decision Trees, Random Forests, SVM, Naive Bayes, KNN, LSVC and Proposed Technique (Ensemble technique of Logistic Regression, Random Forest and Gradient Boosting) to classify the URLs as either phishing or legitimate.
• Evaluation: Evaluate model performance based on accuracy, precision, recall, F1-score, and ROC curve.