ScamGuard: Image-based Identification App for Phishing and Open Attachment Messages Using Variants of Convolutional Neural Networks

The growing prevalence of scam messages in the digital landscape poses a significant threat, particularly in the Philippines where a large proportion of the population is active on social media. These messages are becoming increasingly sophisticated and difficult to distinguish, making the development of effective scam detection techniques essential for protection against scammers. Text-based scam detection models developed in the past years are good at classifying text messages, but are disadvantageous because users may accidentally click on links which redirect to illegitimate sources or malwares. This study discusses the development of an image-based scam detection app powered by Convolutional Neural Networks (CNN). A comparative analysis based on Traditional Learning Approach and Transfer Learning Approach was conducted. Results showed that the Transfer Learning Approach based on Mobile Net V2 architecture yielded an accuracy score of 92% while the Traditional Learning Approach achieved an accuracy score of 91%. The best formulated model was integrated into an app, here we call 'ScamGuard'. The app enables users to upload screenshots of messages and classify them as scam or non-scam. Additionally, the app features a chatbot with predefined responses to address users' queries related to scams. By integrating the complex architecture of a CNN model into an app, convenience is provided to the users for identifying scam messages allowing them to make informed decisions before engaging with potentially scam messages. 

Keywords: Phishing, Scam Detection, Convolutional Neural Networks (CNN), Transfer Learning Models 

To cite: Abejuela, H. J., Lumacad, G. S., Tagupa, J., Salise, K. T., & Gabutan, A. J. D. ScamGuard: Image-based Identification App for Phishing and Open Attachment Messages Using Variants of Convolutional Neural Networks.