Article: Multisource financial data fusion enhanced anomaly transaction detection and early-warning mechanism Journal: International Journal of Information and Communication Technology (IJICT) 2025 Vol.26 No.49 pp.1 - 18 Abstract: This research proposes a CLST architecture that integrates multiple data sources using Siamese neural networks (SNN) to identify unusual financial transactions. By leveraging spatial, temporal, and multimodal feature learning alongside class imbalance handling, the model outperforms existing methods in recall, F1-score, and precision, enabling a robust early- warning system for fraud prevention. Multisource data fusion enhances detection accuracy by combining complementary information from diverse financial streams. While prior studies have applied rule-based, or machine learning methods to unimodal datasets, and recent multimodal approaches show promise, challenges remain in complex financial networks. The proposed hybrid method combines CNNs, LSTMs, MLPs, and SMOTE to address class imbalance, with SNN-based feature extraction improving robustness. Experiments demonstrate maximum precision of 0.937 and an F1-score of 0.787, with SNN + RF and SNN + SVM outperforming traditional and SMOTE-based models. Statistical analysis confirms SNN-based models achieve superior stability and balanced accuracy in anomaly detection. Inderscience Publishers - linking academia, business and industry through research

Title: Multisource financial data fusion enhanced anomaly transaction detection and early-warning mechanism

Authors: Ziwei Rao

Addresses: School of Economics and Management, Wuhan Technical College of Communications, Wuhan, Hubei, 430000, China

Abstract: This research proposes a CLST architecture that integrates multiple data sources using Siamese neural networks (SNN) to identify unusual financial transactions. By leveraging spatial, temporal, and multimodal feature learning alongside class imbalance handling, the model outperforms existing methods in recall, F1-score, and precision, enabling a robust early- warning system for fraud prevention. Multisource data fusion enhances detection accuracy by combining complementary information from diverse financial streams. While prior studies have applied rule-based, or machine learning methods to unimodal datasets, and recent multimodal approaches show promise, challenges remain in complex financial networks. The proposed hybrid method combines CNNs, LSTMs, MLPs, and SMOTE to address class imbalance, with SNN-based feature extraction improving robustness. Experiments demonstrate maximum precision of 0.937 and an F1-score of 0.787, with SNN + RF and SNN + SVM outperforming traditional and SMOTE-based models. Statistical analysis confirms SNN-based models achieve superior stability and balanced accuracy in anomaly detection.

Keywords: multisource financial data fusion; anomaly transaction detection; early-warning mechanism; multimodal learning; fraud detection; credit card transactions.

DOI: 10.1504/IJICT.2025.151062

International Journal of Information and Communication Technology, 2025 Vol.26 No.49, pp.1 - 18

Received: 11 Aug 2025
Accepted: 19 Sep 2025
Published online: 12 Jan 2026 *

Title: Multisource financial data fusion enhanced anomaly transaction detection and early-warning mechanism

Keep up-to-date