Phishing detection technology with a deep learning model on unbalanced training data
Аннотация
Phishing websites refer to an attack where attackers spoof official websites to lure people into accessing to illegally obtain user identity, password, privacy, and even properties. This attack poses a great threat to Internet users and is becoming more and more difficult. Many proposals for identifying phishing websites have proven to be effective and beneficial. To improve the accuracy of phishing by further detection of websites, the article proposes a new convolutional neural network model with an optimal configuration. In particular, the convolutional neural network model first uses a generative adversarial network (GAN) to generate phishing URLs in order to balance datasets of legitimate and phishing URLs. Then it uses the CNN to build a new classifier, which consists of four blocks. Experiments show that the convolutional neural network model achieves CNN accuracy as high as 97.5%, which is higher than that of CNN-LSTM, single CNN and single LSTM by 1.45%, 3.7% and 2.2% respectively.