# Hyderabad Researchers Develop High-Accuracy AI Model for SAP Fraud Detection

> A new AI model developed in Hyderabad combines advanced deep-learning techniques to revolutionize fraud detection in SAP systems, promising enhanced accuracy and dynamic analysis of financial activities.

**Source**: newindianexpress.com | **Published**: 2026-10-11 | **Type**: research

## Key Facts

- New AI model achieves 0.956 ROC-AUC, indicating high accuracy in fraud detection for SAP systems.
- Hybrid approach outperforms traditional methods, revealing vulnerabilities in existing fraud detection.
- 7.1% false-positive rate suggests potential cost savings in auditing processes through improved accuracy.
- Continuous auditing capability indicates strategic shift towards proactive risk management in enterprises.
- Need for real-world testing highlights uncertainty in model reliability, impacting adoption decisions.

## Summary

Researchers from Osmania University and Jawaharlal Nehru Technological University in Hyderabad have developed an innovative hybrid deep-learning model aimed at detecting fraudulent activities within enterprise software systems, particularly SAP S/4HANA. This advancement is significant as it addresses the growing concern of financial irregularities and unauthorized access in complex enterprise environments, which can lead to substantial financial losses and compliance issues for organizations.

The newly proposed framework integrates three advanced deep-learning techniques: a graph neural network for mapping process relationships, a Transformer-based model for analyzing transaction sequences, and a variational autoencoder to learn patterns while managing uncertainty. This multifaceted approach distinguishes itself from traditional monitoring systems that typically depend on static rules and thresholds. By analyzing the dynamic relationships between business processes and transaction patterns over time, the model provides a more nuanced detection capability.

Initial tests on synthetic SAP-related data yielded impressive results, with a receiver operating characteristic area under the curve (ROC-AUC) score of 0.956 and a precision-recall area under the curve (PR-AUC) score of 0.691, alongside a relatively low false-positive rate of 7.1%. These metrics suggest that the model is effective in identifying suspicious transactions and anomalies, outperforming existing comparison models. However, the researchers caution that these results are primarily based on synthetic data and benchmark datasets, indicating a need for further validation in real-world environments.

The implications of this development extend beyond mere detection of fraud. The model’s potential to support continuous auditing, risk management, and regulatory compliance could transform how organizations approach financial oversight. As companies increasingly rely on sophisticated software systems, the ability to detect irregularities in real time becomes crucial. This model could enable businesses to proactively address vulnerabilities, thereby enhancing their overall risk management strategies.

However, while the initial findings are promising, the reliance on synthetic data raises questions about the model's adaptability to large-scale, real-world applications. The researchers have acknowledged the necessity for extensive testing on actual SAP S/4HANA audit logs to confirm the model’s reliability and generalizability. This step is critical for gaining the confidence of enterprises that are hesitant to adopt new technologies without proven efficacy in their specific operational contexts.

The emergence of this AI model signals a broader trend towards integrating advanced analytics and machine learning in enterprise resource planning (ERP) systems. As organizations face increasing regulatory scrutiny and the threat of cyber fraud, the demand for sophisticated detection mechanisms will likely grow. Competitors in the ERP space may need to enhance their offerings to include similar capabilities or risk falling behind in a rapidly evolving market.

Looking ahead, the successful deployment of this hybrid model could catalyze a shift in how enterprises manage compliance and fraud detection. Companies that adopt such advanced technologies may not only improve their operational efficiencies but also gain a competitive edge by fostering greater trust with stakeholders through enhanced transparency and accountability. As the landscape evolves, organizations that invest in innovative solutions like this will be better positioned to navigate the complexities of modern business environments.

## Entities

- **Products**: SAP S/4HANA
- **Technologies**: hybrid deep-learning model, graph neural network, Transformer-based model, variational autoencoder
- **People**: Ram Reddy Jonnalagadda, Kiran Kumar Reddy, Pell Reddy Rajender Reddy, Shshank Chaube, BP Joshi, Manish Kumar
- **Organizations**: Osmania University, Jawaharlal Nehru Technological University, Symbiosis Institute of Technology, Graphic Era Hill University

## Key Concepts

AI model for fraud detection, deep-learning techniques, transaction attributes, continuous auditing, risk management, regulatory compliance, synthetic data testing, real-world deployment

## Definitions

- **hybrid deep-learning model**: A model that combines multiple deep-learning techniques to analyze complex data patterns.
- **ROC-AUC score**: A performance measurement for classification models, indicating the model's ability to distinguish between classes.
- **PR-AUC score**: A metric that evaluates the precision and recall of a model, particularly useful in imbalanced datasets.
- **graph neural network**: A type of neural network designed to process data structured as graphs, capturing relationships between entities.
- **Transformer-based model**: A model architecture that uses self-attention mechanisms to process sequential data, commonly used in natural language processing.

## Use Cases

- detecting suspicious transactions
- unauthorized access detection
- identifying unusual system changes
- supporting continuous auditing
- enhancing risk management
- ensuring regulatory compliance

## Frequently Asked Questions

**What is the purpose of the new AI model developed by Hyderabad universities?**

The AI model aims to detect SAP fraud by identifying suspicious transactions, unauthorized access, and unusual system changes in enterprise software.

**How does the model differ from conventional monitoring systems?**

Unlike conventional systems that rely on predefined rules, the model analyzes relationships between business processes and transaction patterns over time.

**What technologies are used in the AI model?**

The model combines a graph neural network, a Transformer-based model, and a variational autoencoder to effectively analyze transaction data.

**What were the results of the tests conducted on the model?**

The model achieved a ROC-AUC score of 0.956 and a PR-AUC score of 0.691, indicating strong performance in detecting anomalies.

**What are the next steps for validating the model's effectiveness?**

Further testing on large-scale, real-world deployments is needed to establish the model’s reliability and generalizability beyond synthetic datasets.

## Links

- [Read on Welcome.AI](https://welcome.ai/content/hyderabad-researchers-develop-high-accuracy-ai-model-for-sap-fraud-detection)
- [Original source](https://www.newindianexpress.com/cities/hyderabad/2026/Oct/11/hyderabad-universities-find-new-ai-model-to-help-detect-sap-fraud)

---

Source: Welcome.AI | https://welcome.ai/content/hyderabad-researchers-develop-high-accuracy-ai-model-for-sap-fraud-detection