Unsupervised learning for financial statement fraud detection using manta ray foraging based convolutional neural network. (26th September 2022)

• Record Type: Journal Article
• Authors: Singh Yadav, Ajit Kumar; Sora, Marpe
• Is Part Of: Concurrency and computation. Volume 34:Number 27(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-09-26
• Subjects: adaptive density based clustering -- convolutional neural network -- data mining -- deep learning -- financial statement fraud -- manta ray foraging optimization; Parallel processing (Electronic computers) -- Periodicals; Parallel computers -- Periodicals; 004.35
• DOI: 10.1002/cpe.7340
• Languages: English
• ISSNs: 1532-0626

Abstract

Summary: Financial statement fraud detection is a significant topic and a challenging task related to countless applications such as financial sectors, insurance, government agencies, law enforcement and so on. This article presents an unsupervised learning model to identify the financial fraud statements using the hybrid convolutional neural network. The fraud detection scheme can be performed using a few stages: data preprocessing, sampling, feature extraction, feature selection, clustering, and classification methods. The preprocessing and sampling processes are used to clean and recognize duplicate data. Then, the proposed model analyzes both text features and financial variables to provide the best classification of financial statement fraud. In this article, a new fuzzy red deer's algorithm is proposed for selecting the optimal feature set, which can improve the classification accuracy of the fraud prediction model. Subsequently, an adaptive density based clustering (ADBC) approach is introduced for labeling the selected features through the clustering process. Finally, the fraud statements are predicted by proposing a hybrid CNN‐MRFO model wherein the hyperparameters of the CNN model are optimized through the MRFO learning algorithm. The simulation results prove the effectiveness of the proposed model in terms of accuracy (98.85%), precision (98.62%) and F‐measure (99.32%).