Tools
Tools
Shinde, Swati, Shinde, Aditya, Virdee, Bal Singh, Khanna, Ashish and Gaikwad, Abhishek (2026) Unsupervised financial anomaly detection using variational autoencoders and latent clustering. In: Blockchain and Artificial Intelligence for Secure Computer Vision Technologies and Applications. CRC Press, Boca Raton (Florida, USA), pp. 341-356. ISBN 9781003659112
This chapter discusses the application of variational autoencoders (VAEs) to detect outliers in an unsupervised manner from financial datasets. VAEs are employed to learn compact latent representations while preserving key data features. We examine the impact of changing the dimensions of the latent space (1 to 4) and utilization of various normalization methods (StandardScaler and MinMaxScaler) on the identification of anomalies. By utilizing latent space clustering with Mahalanobis distance-based thresholding, our method successfully identifies outliers without having to rely on labels. Experimental results indicate that increased latent dimensions and min–max normalization result in improved anomaly separation and more consistent cluster formation. This chapter highlights the capability of VAE-based approaches to improve fraud analytics by identifying suspicious patterns in financial patterns.
![[thumbnail of A3_Outlier Detection using Variational Autoencoder.pdf]](https://repository.londonmet.ac.uk/style/images/fileicons/text.png "A3_Outlier Detection using Variational Autoencoder.pdf")
Restricted to Repository staff only until 29 July 2027.
Available under License Creative Commons Attribution Non-commercial No Derivatives 4.0.
Download (1MB) | Request a copy
 |
View Item |