An Ensemble Voting Classifier based on Machine Learning Models for Phishing Detection

Abstract

The pervasive threat of phishing attacks has necessitated the development of more effective detection systems. This paper introduces a novel ensemble hard voting classifier that integrates the predictive capabilities of Logistic Regression, Gradient Boosting, and K-Nearest Neighbors for the identification of phishing websites with enhanced accuracy. Our methodology encompasses a comprehensive analysis starting with a rich dataset from Kaggle, consisting of over 11,000 websites, each described by 30 features. Through meticulous exploratory data analysis, we have discerned significant patterns and feature correlations, which informed the subsequent data preprocessing phase. We standardized feature scales using the StandardScaler and split the dataset into an 80-20 ratio for training and testing, ensuring both effective model learning and validation. The ensemble model capitalizes on the diversity of its constituent classifiers, outperforming individual models with an accuracy of 95.02%. Our approach demonstrates that an ensemble hard voting classifier not only improves the detection rate but also provides a balanced precision-recall performance, crucial for real-world applications.