Hybrid Deep Learning with CSHO based Feature Selection Model for Financial Fraud Detection

Abstract

The progressive implementation of AI-related technology for corporate financial risk management has become a priority for banks and other financial institutions. It would appear that financial risk detection models trained on artificial intelligence perform well on fraudulent business recall. As the volume of financial upsurges, the use of traditional machine-learning algorithms for fraud detection is becoming increasingly challenging. Investigation teams may find it extremely challenging to discover safeties fraud from a mountain of electronic evidence without the aid of mechanisation, statistical methodologies, and analytics. Financial fraud can have devastating consequences for a company's stability, as well as significant losses for shareholders, the industry, and even the entire market. Existing fraud detection studies primarily rely on traditional data sources, which use limited information from financial statements. In this paper, we presented a Batch Normalization Based Auto-Encoded Gated Recurrent Unit (BN-AGRU) approach for financial fraud detection that used an auto encoder to capture local features. Through the use of a pre-trained real word vector, we integrated batch normalisation into the AE-GRU model to produce a unique architecture for financial fraud detection. The process of selecting features is handled by the Seahorse Optimizer (SHO). Chaotic Sea- Horse Optimisation (CSHO) is introduced as a hybrid algorithm that strikes a new balance between the exploration and abuse stages.  To mitigate the loss function and capture long-term dependency using the arrangement input approach, we employ lengthy short-term memory as substitutes. Our study adds to existing efforts by expanding on previous modifications to standard GRUs. The experimental findings established that the suggested model outdid state-of-the-art methods across a wide variety of criteria.