Implementasi Convolutional Neural Network (CNN) untuk Deteksi Visual Situs Judi Online

Abstract

Online gambling poses a serious cyber threat in Indonesia, causing financial, social, and psychological impacts on society. According to PPATK's 2025 annual report, the circulation of online gambling funds reached IDR 286.84 trillion throughout 2025, with 12.3 million depositors recorded, despite declining from the previous year due to government intervention. The rapid emergence of new domains renders blacklist-based blocking ineffective. URL and text-content detection methods are also vulnerable to obfuscation, whereas visual characteristics of web pages remain more consistent and harder to manipulate. This study implements a Convolutional Neural Network (CNN) for detecting online gambling sites based on visual analysis of web pages. The novelty lies in a custom CNN architecture with four convolutional blocks for binary visual classification of gambling pages, without relying on URL patterns or text content. The dataset consists of 2,188 web page screenshots collected via Selenium-based crawling, split 70:15:15 for training, validation, and testing. Preprocessing includes resizing images to 216x384 pixels and data augmentation. The CNN architecture employs four convolutional blocks (Conv2D, Batch Normalization, ReLU, and Max Pooling), followed by GlobalAveragePooling2D, Dense layers, and Dropout. Testing achieved 97.27% accuracy with a precision, recall, and F1-score of 0.97, demonstrating the strong potential of this visual CNN approach for automated online gambling site detection.