Optimization-enabled machine learning with feature selection for phishing detection system

Abstract

The rapid evolution of phishing methods has long been a threat to cybersecurity that demands adaptive and intelligent detection techniques. This current work envisions an improved phishing detection system that involves the integration of machine learning (ML) with feature selection for improving the performance of real-time classification. A comprehensive dataset was optimized and preprocessed through Chi-square (CHI) feature selection to reduce dimensionality while preserving the most discriminative features. Random forest (RF) algorithm, selected for its robustness, interpretability, and high computational efficiency for binary classification, was employed and achieved an accuracy of 96.25%, effectively identifying sophisticated phishing attacks. The architecture is designed to operate on large datasets with minimal computational overhead but with high precision and scalability. Experimental results also indicate the dependability of the system with an area under the curve (AUC)-receiver operating characteristic (ROC) score of 1.0, indicating perfect separation of phishing from normal cases. In addition, an adaptive classification model was used to accommodate unknown data sets with a 100% accuracy score. The results above identify the potential of adaptive, feature-optimized ML systems in the ability to handle dynamic phishing attacks and enable proactive cybersecurity defense.