[Paper Review] Year-over-Year Developments in Financial Fraud Detection via Deep Learning: A Systematic Literature Review

Yisong Chen, Conghui Zhao+3 more|ArXiv.org|Jan 31, 2025

Imbalanced Data Classification Techniques5 citations

TL;DR

A Kitchenham-based systematic review of deep learning for financial fraud detection, analyzing 57 studies (2019–2024) across sectors, models, preprocessing, and privacy considerations.

ABSTRACT

This paper systematically reviews advancements in deep learning (DL) techniques for financial fraud detection, a critical issue in the financial sector. Using the Kitchenham systematic literature review approach, 57 studies published between 2019 and 2024 were analyzed. The review highlights the effectiveness of various deep learning models such as Convolutional Neural Networks, Long Short-Term Memory, and transformers across domains such as credit card transactions, insurance claims, and financial statement audits. Performance metrics such as precision, recall, F1-score, and AUC-ROC were evaluated. Key themes explored include the impact of data privacy frameworks and advancements in feature engineering and data preprocessing. The study emphasizes challenges such as imbalanced datasets, model interpretability, and ethical considerations, alongside opportunities for automation and privacy-preserving techniques such as blockchain integration and Principal Component Analysis. By examining trends over the past five years, this review identifies critical gaps and promising directions for advancing DL applications in financial fraud detection, offering actionable insights for researchers and practitioners.

Motivation & Objective

• Assess trends in financial fraud types addressed by DL in recent years.
• Evaluate preprocessing, handling of imbalanced data, and automation improvements enabled by DL.
• Identify advancements in DL models for fraud detection.
• Examine benchmarks and evaluation metrics used across sectors and their relation to imbalanced data.
• Explore how data privacy, anonymization, and regulations influence DL applications in fraud detection.

Proposed method

• Apply Kitchenham systematic review framework to select and synthesize studies.
• Search across PubMed, SSRN, IEEE Xplore, ACM DL, ScienceDirect, and Scopus with domain keywords.
• Screen for original, English-language, peer-reviewed DL-based fraud detection studies from 2019–2024.
• Extract structured data on datasets, models, preprocessing, metrics, and privacy considerations.
• Visualize trends with Python libraries and VOSviewer for keyword networks.

Figure 1: Literature Review Methodology

Experimental results

Research questions

• RQ1 What trends exist in the types of financial fraud addressed using DL in recent years?
• RQ2 How have feature engineering, data preprocessing for imbalanced data, and automation with DL impacted performance and time-to-detection?
• RQ3 What advancements have been made in DL models for financial fraud detection?
• RQ4 What trends exist in benchmarks and evaluation metrics across sectors?
• RQ5 How have data privacy, anonymization, and regulatory rules influenced DL applications for financial fraud detection?

Key findings

• 57 high-quality papers were analyzed from an initial 2,858 results, narrowed to 427 after screening, spanning 2019–2024.
• There was a steady rise in DL fraud detection research from 2019–2021, with a significant increase from 2022 onward and a steep rise 2023–2024.
• Credit card and banking sectors show the largest research activity; crypto/blockchain and payments emerging areas; tax, mortgage/loan, and money laundering are less represented.
• Imbalanced data is a pervasive challenge; 48/57 papers report imbalance, prompting use of SMOTE, GANs/VAEs, ADASYN, and advanced imputations.
• Automation and privacy-preserving techniques (blockchain, federated learning, PCA) are increasingly explored to speed detection and protect data.
• A wide range of models is used, with LSTM, MLP, CNN, Transformer-based methods, GNNs, GANs, and VAEs driving progress; hybrid models and domain-specific metrics are common.

Figure 2: Literature Review Methodology