Optimizing Financial Security: SXI++‐Powered Large Numerical Model for Cross‐Domain Fraud Detection

Manuscript submitted August 25, 2025; accepted September 12, 2025; published November 21, 2025

Abstract—Financial statement fraud remains a major threat, leading to annual global losses exceeding $50 billion and eroding trust in financial systems. This issue is intensified by advanced schemes involving synthetic identities, document forgery, and vulnerabilities in real-time digital onboarding. Traditional rulebased and static machine learning approaches struggle with adaptability, limited features, and high false positive rates. In this study, we employ the Sriya Expert Index (SXI++) within a Large Numerical Model (LNM) to deliver scalable, interpretable, and high-precision fraud detection across various financial domains. Six heterogeneous fraud datasets—bank account, credit card, mobile transactions, cryptocurrency transactions, IEEE-CIS fraud, and simulated fraud—were combined into a unified Master Dataset containing 509,841 records and 394 features. Data preparation included Variational Autoencoder (VAE)-based imputation, categorical encoding, and feature selection. SXI++ aggregated weighted outputs from 5–10 machine learning algorithms into a real-time fraud risk score, leveraging behavioral, transactional, demographic, and temporal patterns. The model was trained on 407,872 records and validated on 102,000 records. Its performance was compared against XGBoost, with accuracy, precision, recall, and AUC as key evaluation metrics. The SXI++ LNM achieved up to 99.58% accuracy, 100% precision, and an AUC of 0.95–0.96, substantially outperforming XGBoost and other baseline models. A strong negative correlation (–0.91) between SXI scores and fraud likelihood established clear classification boundaries (SXI ≤ 0.68 identifying all fraud cases). The SXI++ LNM offers a groundbreaking approach to fraud prevention by integrating multi-source data, real-time scoring, and explainable AI to deliver unmatched predictive performance. Its scalability and interpretability make it suitable for deployment across multiple sectors, enabling financial institutions to proactively reduce fraud risk and maintain customer trust. Future research will examine expansion to other industries and integration with real-time transaction monitoring systems.

keywords—fraud detection, financial transactions, machine learning, SXI++, Large Numerical Model (LNM), XGBoost, Variational Autoencoder (VAE), synthetic identities, real-time scoring, multi-source fraud datasets, explainable AI

Cite: Reeshabh Kumar, Prashant Yadav, Mahesh Banavar, Srinivas Kilambi,"Optimizing Financial Security: SXI++‐Powered Large Numerical Model for Cross‐Domain Fraud Detection," Journal of Advances in Artificial Intelligence, vol. 3, no. 4, pp. 273-286, 2025. doi: 10.18178/JAAI.2025.3.4.273-286

Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).