STAGE framework: A stock dynamic anomaly detection and trend prediction model based on graph attention network and sparse spatiotemporal convolutional network DOI Creative Commons

Ming Wang Shi,

Roznim Mohamad Rasli,

Shir Li Wang

и другие.

PLoS ONE, Год журнала: 2025, Номер 20(3), С. e0318939 - e0318939

Опубликована: Март 17, 2025

As the financial market becomes increasingly complex, stock prediction and anomaly data detection have emerged as crucial tasks in risk management. However, existing methods exhibit significant limitations handling intricate relationships between stocks addressing anomalous data. This paper proposes STAGE framework, which integrates Graph Attention Network (GAT), Variational Autoencoder (VAE), Sparse Spatiotemporal Convolutional (STCN), to enhance accuracy of robustness detection. Experimental results show that complete framework achieved an 85% after 20 training epochs, is 10% 20% higher than models with key algorithms removed. In task, further improved 95%, demonstrating fast convergence stability. offers innovative solution for prediction, adapting complex dynamics real-world markets.

Язык: Английский

STAGE framework: A stock dynamic anomaly detection and trend prediction model based on graph attention network and sparse spatiotemporal convolutional network DOI Creative Commons

Ming Wang Shi,

Roznim Mohamad Rasli,

Shir Li Wang

и другие.

PLoS ONE, Год журнала: 2025, Номер 20(3), С. e0318939 - e0318939

Опубликована: Март 17, 2025

As the financial market becomes increasingly complex, stock prediction and anomaly data detection have emerged as crucial tasks in risk management. However, existing methods exhibit significant limitations handling intricate relationships between stocks addressing anomalous data. This paper proposes STAGE framework, which integrates Graph Attention Network (GAT), Variational Autoencoder (VAE), Sparse Spatiotemporal Convolutional (STCN), to enhance accuracy of robustness detection. Experimental results show that complete framework achieved an 85% after 20 training epochs, is 10% 20% higher than models with key algorithms removed. In task, further improved 95%, demonstrating fast convergence stability. offers innovative solution for prediction, adapting complex dynamics real-world markets.

Язык: Английский

Процитировано

0