Development of a Cybernetics Phishing Detection Approach

Abstract

Phishing being one of the core problems encountered by online community has led to numerous financial loss, identity theft, kidnappings, deaths and lots more. Phishing detection and response software which is a cybersecurity tool that identifies and rectifies phishing threats before the phishing attack causes harm, is a part of the broader threat detection and online security  response. The phishing detection system employed in this study is a one-page web application model aimed at solving phishing problems, with high rate detection accuracy, low rate false alarm and consistent database updates. Machine Learning technology was used by the detection system in extracting and analyzing different features of malicious and phishing URLs.  Extreme Gradient Booster (XGBoost),Decision trees, Multilayer Perceptron and Random Forest models were compared in detecting phishing websites with the aid of Python, Science-kit learn (sklearn) and Flask. The Frontend of the one-page web application was done using  HTML and CSS. The generated dataset from which the models were tested and trained were extracted from various open-source platforms which provided some phishing URLs in various formats like JSON and CSV with hourly updates. Up to 2000 random phishing URLs were collected from this dataset to train and test the Machine Learning models. Out of all these types, the benign URL dataset was considered for this paper. An accuracy of 86.6% with 13.34% false-positive rate was achieved in our proposed approach on our dataset, together with 13.6% false-positive rate and 86.4% accuracy on the benchmark dataset, which performs better than the existing baseline approaches.