HD Pump: A Hybrid Detection Approach for Pump-and-Dump Schemes in Cryptocurrency Exchanges

The adoption of cryptocurrencies has created a favorable environment for price manipulation practices, such as pump-and-dump (PD) schemes. These schemes aim to artificially inflate an asset's price, followed by a rapid sell-off, which may harm unaware investors. Given the brief duration of PD scheme effects, their impact on the asset's price series can be considered anomalies. Most studies rely on classification-based anomaly detection techniques to identify the PD event, which presents an opportunity to explore techniques beyond anomaly detection. To address this, we explore the combination of anomaly and change point detection to enhance pump-and-dump scheme detection. We introduce HD Pump, a hybrid detection method that integrates both techniques. Experimental results demonstrate that our hybrid approach significantly improves performance, achieving a 6.7% increase in precision and a 9.3% increase in recall compared to the benchmark method that solely uses anomaly detection.