Phishing in the Digital Era: A Systematic Review of the Most Promising Detection Techniques (#1124)

Authors

Paucar Cordova, Jhon Alonso

Abstract

Phishing is a growing threat that surpasses traditional detection methods, requiring advanced approaches. The aim of this review was to identify the most promising emerging techniques for phishing detection and assess their effectiveness in overcoming the limitations of traditional methods. A systematic search was conducted using the PICO strategy in academic databases such as Scopus, Redalyc, and SciELO. Out of a total of 293 identified articles, 30 studies published between 2020 and 2024 were selected after applying inclusion and exclusion criteria. Hybrid models in parallel execution (Random Forest, Naive Bayes, CNN, and LSTM) achieved accuracy rates above 99.97%, standing out as the most effective techniques in critical sectors such as finance and corporate environments. These technologies overcome traditional limitations by combining advanced Machine Learning and Deep Learning capabilities. Additionally, vulnerable groups were identified as users of banking and e-commerce services, due to their constant exposure to online transactions; users in governmental and financial sectors, who are key targets because of the sensitive information they handle; and young users (18-25 years old), particularly those with high interaction on social media and limited cybersecurity knowledge. Hybrid models in parallel execution represent a significant advancement in phishing detection, but their effectiveness depends on factors such as the computational load they require and their limited accessibility for everyone. Furthermore, these techniques will be fully effective if complemented with cybersecurity education that fosters user preparedness.