Phishing Websites Detection Based on Hybrid Model of Deep Belief Network and Support Vector Machine

Author:
Open Access
Publisher website
Google Scholar
Abstract
The boosting of financial crimes that employ technical methods has become a critical issue that is urgent to be solved. However, the performance of most of the traditional classification methods are dependent on the quality of the prior knowledge of features. To address these problems, this paper proposed a hybrid model that combines the advantages of deep learning neural network of Deep Belief Network and machine learning method of Support Vector Machines. Firstly, the unidentified URLs from blacklist filtering are processed to have the URLs features extracted, the features are including statistical features, webpage code features and webpage text features. Secondly, deep features are extracted by the quick classification of deep learning model. Lastly, the resulting feature vectors combining with URL statistical features, webpage code features, webpage text features are fed into SVM model for classification. The model was tested on a dataset containing millions of phishing URLs and legitimate URLs, and have achieved the accuracy of 99.96%, the precision rate of 99.94% and the false positive rate of 51.32% which showed better performance than other comparison models.

This publication has 4 references indexed in Scilit: