Phishing Website Detection Based on Data Tuning Methods with PCA of Multidimensional Features by Machine Learning Algorithms

Abstract

Introduction: Recently, especially in the last decade, cyber fraud crimes have increased in various ways to deceive victims, causing financial losses and losses in data and privacy of many users, so many researchers have focused on addressing cyber fraud problems to reduce them.

Objectives: For example, developing a mechanism to detect phishing operations in several ways, including old, traditional or modern methods, which are represented in using artificial intelligence models to detect behaviors or Internet users, whether suspicious or healthy, in addition, some reputable and solid sites have taken it upon themselves to monitor the database by collecting the methods used by hackers, including the famous Al-Mitre website.

Methods: we presented our work to detect whether a website is legitimate or a phishing website using seven algorithms from machine learning algorithms, which are Ada Boost classifier, Decision Tree classifier, Gradient Boosting, KNN algorithm, Logistic Regression, Random Forest classifier and Support Vector Machine.

Results: emply different implementation methods in the form of three experiments, and we concluded that all seven algorithms are almost good, which is Experiment C, in which we benefited from the advantages and disadvantages of previous experiments, and the accuracy obtained from the seven algorithms was as follows: 84.8, 96.6, 83.9, 98.7, 99.6, 99.9, 97.00, respectively.

Conclusions: It is clear from the above results that machine learning algorithms are a good and effective tool in combating malicious activities due to the algorithms' ability to detect phishing sites.