Conditional Dependency-Based Anomaly Detection in ERP Financial Data Using SGKGM And TLFS-DBI-LSTM For Secure Governance

In Enterprise Resource Planning (ERP) systems, ensuring the integrity of financial accounting data is fundamental for significant governance and decision-making. However, previous studies did not focus on analyzing the conditional dependencies between time, user behavior, and operational cycles in ERP financial data, leading to missed operational nuances and inefficient governance. Thus, this paper proposes enhanced anomaly detection in ERP financial data through conditional dependency analysis using sparse Gaussian Kernalized Graphical Model (sGKGM) and Transfer Learned Flamingo and Scaled function based Deep Bi-directional Long Short Term Memory (TLFS-DBi-LSTM). Initially, for missing value imputation and normalization, the custom-made Oracle ERP-based financial data is pre-processed. The features are extracted from the pre-processed data. Then, the extracted features are clustered into user behavior, time, and operational cycle stages. Then, to perform dependency analysis, the proposed sGKGM is used, and then the dependency features are extracted. Now, the similarity is analyzed between the dependency features. Here, the extracted features and similarity score are fed into the proposed TLFS-DBi-LSTM, where the ERP financial data is classified into abnormal and normal, followed by deep explainability. The severity assessment is done for the abnormal data regarding the severity score. Thus, for decision-making, the alert is sent to the admin. Lastly, the financial data and governance policy are preserved and then securely stored in the cloud. Therefore, the proposed approach had superior performance with 99.6% accuracy.