Feature Engineering Approaches for Predictive Modeling of Phishing Campaign Effectiveness

Abstract

Phishing attacks continue to evolve in sophistication, targeting diverse industry sectors with varying degrees of effectiveness. This study investigates comprehensive feature engineering methodologies for developing predictive models that assess phishing campaign effectiveness across multiple dimensions. Through systematic analysis of industry-specific attack patterns, URL-based lexical features, and multi-layered detection approaches, we propose an integrated framework that combines traditional heuristic methods with advanced machine learning techniques. Our methodology leverages feature selection algorithms applied to the ISCX-URL2016 dataset comprising 9,964 phishing URLs and 10,000 legitimate URLs, identifying nine critical features that demonstrate strong discriminative power in predicting campaign success rates. Analysis reveals that financial services, Software as a Service platforms, and webmail systems constitute the primary targets, accounting for sixty percent of phishing campaigns. The multi-layered detection framework integrating list-based, visual similarity, and heuristic machine learning approaches achieves superior performance through optimal feature engineering. This research contributes actionable insights for prioritizing defensive strategies based on industry vulnerability profiles and predictive feature importance rankings.