Stepping up Phishing Detection using Convolutional Neural Network Algorithm

Abstract

Phishing websites continue to pose a significant cybersecurity threat by deceiving users into disclosing sensitive information, making accurate and reliable detection methods increasingly important. This paper evaluates the effectiveness of several machine learning algorithms for phishing website detection, including Convolutional Neural Network (CNN), Support Vector Machine (SVM), Decision Tree, K-Nearest Neighbor (KNN), Gaussian Naïve Bayes (GNB), and Gradient Boosting (GBoost). The proposed approach first analyzes the CNN learning process using training and validation accuracy and loss curves to assess model convergence and generalization capability. Subsequently, the performance of all classifiers is evaluated using confusion matrices, accuracy, precision, recall, and F1-score. Experimental results show that the CNN achieves the highest overall accuracy of 93%, with balanced precision, recall, and F1-score of 0.93 for both legitimate and phishing website classes. SVM, KNN, and GBoost also demonstrate competitive performance with 91% accuracy, whereas the Decision Tree achieves 89% accuracy. In contrast, Gaussian Naïve Bayes produces the weakest performance with an accuracy of only 51%, indicating that its feature independence assumption is inadequate for modeling the complex characteristics of phishing websites. These findings demonstrate that deep learning, particularly CNN, provides more accurate and balanced phishing detection than conventional machine learning methods. The proposed approach offers an effective solution for practical phishing detection and has the potential to strengthen cybersecurity systems by reducing both false positives and false negatives. Future work will investigate advanced deep learning architectures, explainable artificial intelligence techniques, and larger real-world phishing datasets to further improve detection accuracy, robustness, and model interpretability.