Improving Fraud Detection Using Machine Learning Models

Due to the evolving nature of risk, legacy fraud detection techniques -- typically rules-based systems -- have a difficult time scaling to meet new fraud patterns while achieving high precision. Machine learning systems have several inherent advantages, among them being able to continually learn and adjust, as well as being able to support extremely large datasets and detect patterns which humans cannot. In this presentation, Simility shares recommendations concerning labels, feature engineering, and feedback, with insights on which models and algorithms are best suited to specific scenarios. The review concludes with four areas in which machine learning is evolving and how machine learning systems are being adapted to meet new challenges.

Note that this presentation is also available as a webinar. To view the webinar, click here.