Downloads: 0 | Views: 37
Informative Article | Engineering Science | India | Volume 10 Issue 9, September 2021 | Rating: 4.9 / 10
A Comprehensive Review on Generative Models for Anomaly Detection in Financial Data
Abstract: Anomaly detection in the financial industry is one of the most critical aspects of identifying fraudulent transactions, outliers, and other uncommon patterns that might suggest some illegal action. Traditional techniques need help catching complicated financial data abnormalities with high dimensionality, often delivering incorrect results. Introducing more advanced generative modeling techniques that can learn the unique underlying distribution of average financial data to analysis acts as a more reliable approach. Through mimicking usual data points, these models generate random values when parameterized with a given set of characteristics; thereby, these become anomalies whose probability of being caused by the learned distribution is extremely low. This review delivers a detailed and accurate description of the state - of - the - art innovative generative models used in financial anomaly detection, such as autoencoders, GANs, normalizing flows, energy - based models, and more. The strengths and weaknesses of the given approaches are appraised. The following paragraph discusses the challenges embedded in frameworks for creating and deploying generative models, including interpretability, the protection of privacy, working with multidimensional data, and scarcity of data. Generative models can enhance the accuracy of anomaly prediction. However, researchers must adapt the algorithms to anomaly forecasts in complex market data. Targeted measures to solve the imperfections of generative models will fully unravel the potential ideal contemporary detection system that ensures lower false positives and false negative rates. Many financial institutions already employ powerful machine learning models to address financial crimes, such as fraud, money laundering, and insider trading. Still, turning it from academic promise to practical use could reduce most of these significantly.
Keywords: anomaly detection, generative models, financial data, autoencoders, GANs, normalizing flows
Edition: Volume 10 Issue 9, September 2021,
Pages: 1767 - 1770