فائزه موسوي فيروزكلا

Pump-and-dump schemes are among the most preva‎lent and widely exploited forms of financial fraud, particularly in cryptocurrency markets. This type of fraud involves the artificial manipulation of asset prices, creating sudden market fluctuations aimed at generating quick profits for a specific group of individuals. Recently, the detection and prediction of such activities have become a major research focus in financial market security, as timely identification can help prevent financial losses and contribute to a more transparent and secure market. Most existing studies in this area have primarily concentrated on predicting and detecting the occurrence of pump-and-dump events. However, in contrast to this approach, significantly less attention has been given to the coin prediction on cryptocurrency—predicting which digital asset is likely to be pumped within a specific timeframe. Given its critical importance in financial analysis and investment decision-making, this issue warrants further exploration. To address this challenge, we collected data from Telegram channels associated with pump-and-dump activities over the period from 2022 to 2024. Subsequently, we introduced a novel labeling method based on textual and market data analysis, which led to the identification of 1,200 pump events within this timeframe. This approach effectively refined the dataset for more precise analysis and target cryptocurrency prediction. To uncover hidden patterns within sequential data, we employed attention mechanisms and transformer models. These methods successfully captured complex relationships within the data and provided highly accurate predictions for target cryptocurrencies. Based on the HR@K eva‎luation metric across multiple values of kk, our results demonstrated an approximate 10% improvement compared to prior studies in this domain. Ultimately, our model achieved an AUC score of 89%. Our findings underscore the significant role of integrating both textual and financial market data in enhancing the accuracy of pump-and-dump event predictions.

بازارهاي مالي , پامپ و دامپ , تحليل داده‌هاي متني , افزايش مصنوعي , دستكاري بازار

Financial Markets , Pump-and-Dump , Textual Data Analysis , Artificial Inflation , Market Manipulation

Faezeh Mousavi Firoozkola

Dr. Reza Entezari-Maleki