Uncovering coin market manipulation on X using machine learning
Makaleler > Tam Makale · 2026
Yazarlar: Zehra Nur ÖZBAY, Aybars UĞUR, Mehmet Emin DALKILIÇ
Dicle Üniversitesi Mühendislik Fakültesi Mühendislik Dergisi, Uluslararası, 2026, cilt 17
Özet
Spam-bots have become a widespread problem due to the ease of creating accounts on the X platform. These bots are designed to automatically generate and disseminate unsolicited content, frequently for advertising purposes or events of questionable legality. A key challenge in detecting these spam-bots is their ability to create accounts rapidly, post certain volumes of content around an event, and then become dormant. This study examines spam-bot accounts created to promote cryptocurrency trading on X. We present a novel approach to obtain a spam-bot dataset for coin-related tweets, including data collection, analysis, and labeling. For data preprocessing, we applied Natural Language Processing techniques, which resulted in an imbalanced dataset. We used BERT embeddings for feature extraction and evaluated the performance of nine machine learning algorithms in detecting the tweets of spam-bot accounts. The SVM model achieves the highest cross-validated accuracy of 98.09% on the balanced dataset, produced using the SMOTE oversampling technique. On the original imbalanced dataset, Logistic Regression has the highest accuracy at 96.46%. The cross-validated accuracy of SVM is further improved through hyperparameter tuning, resulting in 99.70%.