Fraudulent E-Commerce Website Detection Using Convolutional Neural Network Based on Image Features

Due to the growth of e-commerce’s website, attackers have motivated to intrude these e-commerce websites. These illegitimate websites, also known as fraudulent websites, exploited the Internet to trick people into committing fraud or engaging in malicious cyberattacks, which cost billions of dollars of loss. Numerous research and methodologies have been developed to recognize fraudulent Internet websites, yet none of them has been able to offer a feasible solution to bring an end to these illegal acts. Moreover, as the number of fraudulent webpages increases, so does the sophistication of the threat. Due to this issue, this paper presents a method for detecting fraudulent websites that is based on image of the website that may contain some patterns that show it is fraudulent. In this paper, convolutional neural network (CNN) classification is used to classify the fraud and legal website image utilized. A total of 530 website images were collected with 272 fraud website images and 258 fraud website images. The websites are partitioned into 80% (370 images) samples as training set, 10% (80 images) samples as testing set, and the rest 10% (80 images) samples as validation set. Three CNN optimizers namely including Adam, Root Mean Square (RMSprop), and Stochastic Gradient Descent (SGD) were tested for training and validation accuracy and loss in different epoch. The results show that proposed fraudulent website detection model accuracy of 78.25% using Adam optimizer for 120 epochs.