Enhanced Phishing Website Categorization Using Random Forest with Sea Horse and Jellyfish Search Optimization

Abstract

In contemporary society, with advancements in science and technology, many global activities, ranging from financial transactions to information transfers, are conducted through the Internet via dedicated websites and applications. Unfortunately, the prevalence of online platforms has increased the proliferation of fake websites aimed at exploiting sensitive data, such as bank card information and personal details. It addresses the problem of cybersecurity w.r.t. the categorization of a set of 1353 websites by a machine learning algorithm into three categories, namely phishing, suspicious, and legitimate URLs. The dataset was gathered from published papers and divided into 70-30 in the training and testing phases. This will help keep members' banking and personal data much safer online. This paper uses the RFC model with two optimization schemes, Sea Horse Optimizer (SHO) and Jellyfish Search Optimization Algorithm (JSOA), to improve performance. After that, optimized versions of the schemes are tagged as RFSH and RFJS, respectively. After extensive training and testing on these three schemes, the best model was identified by comparing the performances of the three on the database in hand. The RFSH model performed better predicting, achieving 0.952 for all the data. It outperformed the RFJS model with a precision of 0.932 and the RFC single framework with an accuracy of 0.9106. Hence, it emerged as the best-predicting model.