Скачать или смотреть IGI050 ScamSeal Real Time Scam Detection on Social Media Using Hybrid Machine Learning 2

The rapid expansion of major social media platforms, including Instagram and Facebook, has fueled a sharp increase in sophisticated scams, exposing millions of users to financial and privacy risks. Existing detection systems lack the real-time accuracy and adaptability needed to provide effective protection. This research introduces ScamSeal, a comprehensive system employing a bilingual hybrid machine learning approach to address these challenges by classifying fraudulent social media content in real-time in both English and Malay. The research explored 15 models, including 3 baseline classifiers, Random Forest (RF), Decision Tree (DT), and Support Vector Machine (SVM), and 12 hybrid models created through pair-wise and tri-model combinations. These models were trained using ensemble methods such as stacking, bagging, and boosting to determine the optimal configuration for cross-platform scam detection. Hybrid models demonstrated superior performance. For Instagram, Bagging DT-SVM outperformed other models by the result of 92.76% accuracy in classifying scam comments with English dataset, while Bagging RF-DT-SVM model obtained 90.36% accuracy for Malay. Meanwhile, for Facebook, the results of Bagging RF-SVM achieved highest accuracy, which is 84.93% in detecting fraudulent posts for English content, whereas Stacking RF-SVM proved optimal with 84.71% accuracy for Malay dataset. These results validate the capability of hybrid machine learning in capturing diverse fraud patterns across platforms. To provide immediate and practical protection, the ScamSeal system is deployed as a functional Google Chrome browser extension. This confirms the approach's suitability for computationally efficient, real-time anti-scam applications that enhance user safety on social media.