Approaches to Fraud Detection: Autoencoder, Isolation Forest and More
Summary
“Experts estimate federal government losses from potential fraud at nearly $150 billion” (source) and new data shows the federal trade commission received 2.8 million reports of fraud in 2021 from consumers (source). In the first part of this webinar, we will work with labelled data to perform classical machine-learning approaches to fraud detection such as the random forest. Then we will cover a deep learning technique, the autoencoder, to find fraudulent data points. In the second part of the webinar, we will focus on data without labels of fraudulent activity using visualisations, classical statistics, and machine learning. You will learn how easy it is to generate multiple visualisations, perform statistical analysis, and use two machine learning algorithms – Isolation Forest and DBSCAN – all to detect fraudulent activity in the free, open-source KNIME Analytics Platform.