Phishing website detection for e-banking by inclined planes optimization algorithm

Abstract

This paper presents a novel approach for detecting phishing websites in internet banking by utilizing the Gravitational Search Algorithm (GSA) for feature selection and classification. The authors highlight that while various methods exist for phishing detection, many ignore the short lifespan of phishing sites and the need for reduced computational complexity. To address this, they propose a rule-based feature selection mechanism that dynamically identifies the most relevant characteristics of phishing websites based on defined thresholds, thereby optimizing the detection process.

The proposed method operates in three main stages: feature extraction, feature selection, and classification using GSA. Features are extracted from website URLs, content, and structure, and are then filtered using a threshold-based rule set to retain only the most discriminative attributes. This selective approach significantly reduces the number of features needed—from an initial set of 15 down to as few as 3 or 8—without compromising detection accuracy. The use of GSA, inspired by the laws of gravity and motion, enables efficient and rapid convergence toward an optimal classification model, avoiding local optima and minimizing computational overhead.

Key achievements of the research include a high phishing detection accuracy of up to 95.12% when using 15 features, with performance remaining strong even with fewer features. The method also demonstrates low false positive and false negative rates, along with reduced processing time per website compared to other meta-heuristic approaches such as PSO, ACO, BFOA, and MBAT. This balance of accuracy, speed, and low resource consumption makes the approach particularly suitable for real-time phishing detection in banking environments.

Overall, the study successfully introduces a lightweight, adaptive, and effective phishing detection system that leverages intelligent feature selection and the gravitational search algorithm. The results validate the feasibility of using fewer, well-chosen features to maintain high detection performance while lowering computational demands—a significant step forward in the ongoing effort to secure online banking against evolving phishing threats.