EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Research on high-frequency financial transaction behavior recognition and prediction method integrating machine learning

Yuqi li

Data and Metadata · 09.05.2026 · 5 · 1459

DOI: 10.56294/dm20261459 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Data and Metadata, online erschienen am 09.05.2026. Verfasst von Yuqi li. Die Arbeit nennt 24 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Introduction

Existing high-frequency trading recognition systems mainly focus on price prediction or rule-based surveillance, but they cannot sufficiently integrate transaction sequences, order book structures and account-level relational behavior under low-latency constraints.

Objective

This study proposes a machine learning-integrated framework for high-frequency financial transaction behavior recognition, short-horizon prediction and anomaly warning.

Method

The framework constructs multimodal features from tick-by-tick transaction records, limit order book snapshots and account behavior logs. A CNN-LSTM-GNN fusion model is designed to capture local microstructure patterns, temporal dependencies and account interaction structures. Bayesian optimization is introduced for hyperparameter selection, while anomaly warning is implemented through order-book distortion features, account behavior profiles and reconstruction-based abnormality scores.

Results

Out-of-sample experiments show that the optimized fusion model achieves 95.1% behavior recognition accuracy and 94.4% macro-F1. For 10 ms price prediction, the model obtains an RMSE of 0.013 and a directional accuracy of 67.8%. In anomaly detection, the recall reaches 92.5% and the AUC reaches 0.967. The system also achieves 9.2 ms latency and a throughput of 50,000 records per second.

Conclusions

The results indicate that multimodal feature fusion and account-level graph modeling can significantly improve HFT behavior recognition and abnormal trading warning. The proposed framework provides a practical technical path for real-time financial market surveillance and risk-sensitive trading monitoring.

Zitieren

APA

li, Y. (2026). Research on high-frequency financial transaction behavior recognition and prediction method integrating machine learning. Data and Metadata, 5, 1459. https://doi.org/10.56294/dm20261459

Vancouver

li Y. Research on high-frequency financial transaction behavior recognition and prediction method integrating machine learning. Data and Metadata. 2026;5:1459. doi:10.56294/dm20261459

BibTeX
@article{li2026,
  author = {li, Yuqi},
  title = {Research on high-frequency financial transaction behavior recognition and prediction method integrating machine learning},
  journal = {Data and Metadata},
  year = {2026},
  volume = {5},
  pages = {1459},
  doi = {10.56294/dm20261459}
}
Kostenlos starten →