Visualisation and Classification of Phishing URL using Ensemble Learning Algorithms and Hyper-Parameter Tuning

The necessity to concentrate on internet security has arisen due to technological and internet advancement. The primary objective of cyberattacks is to obtain sensitive data, including names, passwords, and other details. Cyberattacks are in practice in many ways, and phishing is one of the types. It involves tricking people into divulging sensitive data like location, passwords or credit card details by disguising it as a respected organization or a helpful resource. According to a report from SlashNext, phishing attacks rose by 61% in 2022. Phishing attacks can be employed in various ways and frequently start with an email or developing a hostile website, or redirecting the user to a spam on malicious site. This research paper aims to identify abnormalities in the phishing URLs from visualisation of the attributes and also classify the URLs using binary classification. The binary classification involves two classes: phishing URLsand legitimate URLs. In order to achieve this task, we first investigated different features of the URL using data visualisation and performed data transformation and feature selection based on correlation analysis and F-test to understand the contribution of the features. The visualisations attempted in this research are useful for finding the visible difference in the abnormal activity done by the attackers. Machine and Ensemble learning methods were applied once feature selection for classification of the URLs were initiated. This was followed by hyper-parameter tuning on the best performing model. The output of our experiments shows that LightGBM is giving better accuracy of 98.4%.