Malicious Website Prediction Using Machine Learning Methodologies

Abstract

Nowadays, digital technology has advanced faster than any previous invention, leading to widespread use of machine learning algorithms for generating predictions or decisions without explicit programming. These algorithms rely on a sample set of data, known as training data, to function effectively. However, the absence of high-quality data poses a significant challenge in machine learning, as data quality is crucial for the algorithm's performance. In phishing detection, the ultimate accuracy depends on various key features, including the URL, domain identity, security, and encryption criteria. To extract and verify these criteria from phishing data sets, we utilize regression techniques and classification algorithms. Specifically, we employ decision tree and logistic regression methods as two machine learning techniques. Logistic regression, a standard approach for binary classification problems, originates from the statistical discipline and achieves a 95% accuracy rate on trained data sets. Decision trees, a form of supervised machine learning, continuously split data based on specific parameters and consist of decision nodes and leaves, representing choices and outcomes, respectively. Decision trees achieve an 85% accuracy rate on trained data.