High-frequency trading (HFT) has emerged as a major force in financial markets, with algorithms executing orders in milliseconds, significantly influencing market dynamics and price discovery. Understanding HFT strategies is critical for maintaining market stability and ensuring effective regulatory practices. This research focuses on modeling HFT behavior using Limit Order Book (LOB) data, a powerful tool for capturing detailed real-time market activity. An unsupervised learning framework is introduced, using an LSTM autoencoder to identify temporal patterns in auction market data. KMeans clustering is applied to categorize distinct trading behaviors, providing valuable insights into trader strategies. The FI-2010 dataset is utilized to validate the methodology, demonstrating its potential in areas such as trader profiling, detecting market anomalies, and offering regulatory insights. The approach does not require labeled data, making it highly scalable and effective for large-scale market analysis. This method shows promise for real-time market surveillance and improving financial market transparency.
Song Yang (Tue,) studied this question.