PHISHING WEBSITE DETECTION USING EXTREME LEARNING MACHINE

• Main Author: Lazuardi Yusuf, Hisham
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/42157
• Institution: Institut Teknologi Bandung

The increasing of internet users in the past few years causing the frequency of cybercrime especially phishing to increase rapidly. Phishing is a social engineering attack that deceives the victim into giving a valuable personal information to the phisher or attacker. Phishing is one of the most common cybercrime that attacks internet users and have caused many harm to internet users. The increase in phishing attack frequency has led researchers to develop new methods to prevent and detect phishing attack. With many frequency of attacks and losses caused by phishing attack, an alternative solution for phishing detection is needed by utilizing machine learning using Extreme Learning Machine algorithm. Extreme Learning Machine is a supervised learning algorithm with feedforward neural networks that can be used for classification, regression, or feature selection. In this final project, Extreme Learning Machine is used to detect phishing website with feature selection or dimensionality reduction using four algorithms namely Principal Component Analysis, Correlation Feature Selection, Genetic Algorithm, and Forward-Backward Selection. Phishing website dataset used in experiment in this final project contains 11.055 data and 30 features from website that indicate phishing properties. The results from phishing website detection using test data with Extreme Learning Machine yields an accuracy score of 95,30% and recall score of 94,30%. From the results using test data, Extreme Learning Machine combined with feature selection or dimensionality reduction techniques produce better performance in phishing website detection than without feature selection in general. Extreme Learning Machine also gives pretty good performance in detecting phishing website because it can produce accuracy and recall score close to Support Vector Machine algorithm which is used as an upper baseline.