Screenshot-Based Classification of Online Gambling Websites with MobileNetV2

Abstract

This study aimed to develop and evaluate an automated system for classifying online gambling websites from screenshots by integrating image processing with deep learning via the MobileNetV2 architecture. This convolutional neural network, pre-trained on the ImageNet dataset, was implemented using transfer learning to enhance performance on domain-specific data. The experimental dataset consisted of 1,200 screenshots, equally divided between gambling and general websites. Data preparation involved standard resizing and normalization, with the dataset partitioned into train-validation-test sets at ratios of 70:20:10 and 80:10:10 for empirical comparison. Key parameters, including batch sizes of 8 and 16 and training epochs ranging from 10 to 50, were adjusted alongside the application of early stopping to prevent overfitting and ensure model stability. Performance was evaluated using accuracy, precision, recall, F1-score, and ROC AUC to reflect comprehensive classification capabilities. The results demonstrated that an 80:10:10 data split combined with a batch size of 8 and 50 epochs yielded the highest performance, achieving an accuracy of 0.97, precision of 0.98, recall of 0.95, F1-score of 0.95, and an excellent ROC AUC of 0.9969. The findings indicated that MobileNetV2 was highly effective in accurately and stably identifying gambling websites. Due to its compact size and low computational requirements, the model is well-suited for real-world monitoring systems and law enforcement support. However, the study was limited by its dataset size and binary classification scope. Future research should increase data diversity, compare alternative architectures, and develop multi-class classification to expand applicability in more complex contexts.