Financial Fraud Detection Using AEO-DMOA Based 1D-FRCNN Model with Effective Feature Selection Technique

As a global issue, financial fraud has severely hampered the steady expansion of financial markets. However, if the ratio of legitimate businesses to fraudulent ones is particularly large, it might be difficult to spot fraud in a dataset. Therefore, solutions have been created to help stakeholders make better decisions through the use of intelligent financial statement fraud detection. However, most existing methods primarily take into account the numerical data found in financial statement ratios, while textual data, notably Chinese-language remarks on the subject, has been underutilized for classification. Therefore, the purpose of this study is to create a model-based scheme for financial fraud detection. In this study, the best features from the preprocessed data are chosen using Hybrid Enhanced Glowworm Swarm Optimization (HEGSO). Hybrid optimization (AEO-DMOA) is then used to regulate the best values for the network's hyper-parameters, including its learning rate, epochs, momentum, and batch sizes, before a one-dimensional faster region-based convolutional neural network (1D-FRCNN) is recommended for classification. Dwarf Mongoose Optimization Algorithm (DMOA) and Artificial Ecosystem-based Optimization (AEO) are combined in this model to create an effective algorithm that strikes a better balance between exploration and exploitation. Accuracy, precision, recall, and f1-score were optimized to 99.10%, 98%, 96%, and 97%, respectively, as shown by the assessment of the study effort. For fraud detection challenges, the suggested model performs improved than state-of-the-art learning procedures