Phishing Website Detection Using Effective Classifiers and Feature Selection Techniques

Phishing is a relatively new form of network assault where a web page illegally invokes current users to request financial or personal data or passwords. This act jeopardizes the privacy of many users and consequently, ongoing research has been carried out to find detection tools and to develop existing solutions. Classifiers based on machine learning can be used to detect phishing websites effectively and therefore, various machine learning classification algorithms i.e. Naive Bayes, J48 and HNB are implemented and compared through this research. In addition, performance of a classifier combining HNB and J48 was also closely observed as a solution to the stated problem. The study proposes a novel manual feature selection approach and presents a comparative study with Filter method feature selection techniques. The dataset used in this research is collected from UCI machine learning repository, has 2670 instances and 30 attributes of website structure. The empirical result indicated that the Address bar based feature group achieved the highest accuracy in detecting phishing website. In addition, two top algorithms, HNB and J48, were developed for an integrated multi-classified process. The findings have shown that combining techniques results in 96.25% accuracy in the identification of phishing websites for all apps.