Phishing Website Detection Based on Multidimensional Features Driven by Deep Learning: Integrating Stacked Autoencoder and SVM

Abstract

Background Phishing attacks on financial institutions are increasing, demanding improved detection systems. Fake websites pose major risks to consumers, necessitating precise, automated techniques of detection to minimize escalating cyber threats. Methods A deep-learning model that combines a stacked autoencoder for dimensionality reduction and noise filtering with a Support Vector Machine (SVM) classifier is proposed. This hybrid model examines multidimensional data from websites to detect phishing attempts. Objectives Create a robust phishing detection system employing sophisticated deep-learning algorithms, to achieve improved accuracy, precision, and lower false-positive rates in real-time settings. Results The hybrid model showed increased precision, accuracy, and fewer false positives. Performance measures such as AUROC confirmed the model's ability to detect phishing websites, beating traditional methodologies. Conclusion The suggested system provides a scalable, cost-effective solution for phishing detection, hence boosting financial institutions' security. Its performance demonstrates potential for combating evolving cyber threats with greater precision and reliability.