Stacking ensemble classifier for phishing detection: A cyber security model with efficient feature descriptor

Abstract

Given the dynamic nature of cyber threats, phishing attack detection is still a significant difficulty in the field of cyber security. This paper suggested a novel method for detecting the phishing attack through the ensemble classifier coupled with an efficient feature descriptor, involves preprocessing, extraction of features, and classification steps. At first, the input data has been preprocessed using the proposed normalization techniques to ensure standardized and consistent data representation. Subsequently, features like improved entropy, statistical attributes and mutual information are obtained from the preprocessed data, capturing the characteristics of phishing attempts. These extracted features serve as inputs to a stacked ensemble classifier made up of Support Vector Machine (SVM), Neural Network (NN) Random Forest (RF) and improved Long Short-Term Memory (LSTM), where the outputs of SVM, NN and RF classifiers are combined and fed into an improved LSTM classifier. The improved LSTM model leverages temporal dependencies to provide the final classification output, distinguishing between “attack” and “non-attack” instances. Through experimentation on benchmark datasets, the proposed approach demonstrates superior performance with respect to precision, F-measure, and recall, showcasing its efficacy in phishing detection. This research contributes to advancing cybersecurity by providing a resilient and comprehensive solution for detecting the phishing threats in digital environments.