Intro: Fraud detection in tabular fraud-analytics dataset
Used DL/CV Techniques: VAE(Variational Autoencoder), K-Means Clustering, Isolation forest, Elbow-method for K-Means
Tech-stack used: Pytorch and Python
Metrics: If the cluster size is small or the cluster are at boundary are consider as fraud
What is done during the project: Find out no.of fraud transactions in given dataset
Dataset used: check out report and code
Given dataset we reduce dimension by VAE and perform K-Means on reduced latent space and then through the metrics assumed we found 37 anomalies were present and that matches with isolation forest unsuperised method which we have done additionaly to validate our method was correct.
For more info check out code and report.