Feature selection for efficient classification of phishing website dataset

The Internet is gradually becoming a necessary and important tool in everyday life. However, Internet users might have poor security for different kinds of web threats, which may lead to financial loss or clients lacking trust in online trading and banking. Phishing is described as a skill of impersonating a trusted website aiming to obtain private and secret information such as a user name and password or social security and credit card number. However, there is no single solution that can prevent most phishing attacks. For phishing attacks, various methods are required. In this thesis, a feature selection method and the Navie Bayes classifier are presented for the phishing Websites dataset. In this study, phishing dataset retrieved from UCI machine learning repository is used. This dataset consists of 11055 records and 31 features. The research presented in this thesis aims at reducing the number of features of the used dataset as well as obtaining the best classification performance. Feature selection algorithms are used to reduce the dataset features and to obtain a high system performance. In addition, the performance of feature selection algorithms is compared using the Naive Bayes classifier. Finally, a comparative performance in reducing dataset features using the common classification algorithms is given. The results show that an effective phishing detection can be made with feature selection that reduces the dataset features.