Pengembangan Chrome Extension untuk Mengidentifikasi Phishing Website berdasarkan URL dengan Algoritma Random Forest

Abstract

The ever developing technology makes internet one of the most important part in human’s daily activity. This development is also followed by the increase of phishing activity which is not only in quantity, but also in the variety of techniques. The loss caused by phishing attacks is quite big. There are a lot of applications for preventing phishing attacks, but most of them are still not accurate enough. Several studies show that ensemble learning algorithm has a good capability in detecting phishing website.
In this research a chrome extension which uses a Random Forest model to detect phishing websites has been developed. Random Forest is one of the most well-known ensemble learning algorithm. The most important hyperparameters which would be experimented with are n_estimators, min_samples_leaf, min_samples_split, max_features, and max_depth. Features used are Lexical features which are based on references from other researches, and Domain-based features which are the newly proposed ones, comprised of Global Page Rank, Average Daily Time, Sites Linking In, Domain Age, and Registration Period. All features are obtained only from the URL.
This research shows that dataset quality is the most impacting factor in making a good model. Hyperparameter tuning is also an important part but is only limited to certain scenario. The newly proposed features could make an improvement to the model’s performance. From several experiments, the usage of Lexical and Domain-based features has successfully achieved the best accuracy of 98.28%.