Enterprise digital transformation faces a critical challenge in financial fraud detection which can be solved through precise identification of fraudulent activities with advanced ML and DL approaches. The Paper establishes an Optimized CNN-LSTM Model with Attention and Graph Neural Network Integration designed to improve financial fraud detection through combined spatial and temporal patterns evaluation of financial transactions. Feature extraction through CNN-LSTM networks detects sequential dependencies while an Attention Mechanism selects crucial features from input data. The Graph Neural Network works alongside the proposed system to evaluate financial-entity connections which improves detection efficiency. Semiconductor and statistical methods are applied to the model as Principal Component Analysis performs feature extraction and missing data is replaced with mean imputation. Training data for the model comes from the Finance Fraud Detection Dataset at Kaggle which presents both financial ratios in a structured format and financial report text in unstructured form. The proposed system reaches an accuracy level of 96.0% while surpassing traditional models with 85.3% Logistic Regression and 89.5% Random Forest and 91.2% XGBoost and 92.5% standard LSTM. Python uses the platform of AWS SageMaker to execute the model to optimize the computational processes. The results of the experiments suggest that the hybrid CNN-LSTM based on Attention and GNN is an effective method of detecting fraudulent financial transactions by using deep learning and network analysis. The studies examine the responses of model performance to certain implementation practices of feature scaling and data preprocessing. Future research on interpretability and real-time scalability on fraud detection will be beneficial to enterprise applications.
Chaluvadi et al. (Wed,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: