Coinbase Logo

Using advanced Machine Learning models to protect our users accounts

TL;DR: Our Machine Learning team is committed to developing sophisticated modeling techniques to detect malicious actors in order to keep our users’ accounts safe and secure. Besides sizable commitment in Data and Machine Learning, we continue to invest in our Security, Risk and Machine Learning Platform teams so that they have the foremost tools and resources necessary to be proactive against fraudsters.

By Rajarshi Gupta, Senior Director, Head of Machine Learning and Kenneth Dai, Engineering Manager, Machine Learning Risk


, October 12, 2022


For Cybersecurity Awareness Month, we wanted to highlight the work of our Machine Learning team. At Coinbase, one of our core values has always been to provide a safe and secure platform for our users to participate in the crypto economy. We believe that our community needs access to industry-leading security features that are easy and convenient to use. Our commitment to all our users has not changed, which is why we continue to innovate in order to provide the safest, most secure experience possible. 

In the last two decades, machine learning has greatly impacted the daily lives of countless people around the world. Machine learning technology is now used across many industries and disciplines, in everything from agriculture and economics, to athletics and the arts. Machine learning models  improve the lives of people in medical diagnostics, email filters, speech recognition, computer vision, and much more.

Our Data, Risk, and Security teams partnered to develop and refine machine learning models to use on our platform.  These models make our platform more secure, help prevent bad actors from causing harm, and do so without negatively impacting the user experience for our customers. The Mission is to keep building scalable, adaptive, blockchain aware ML systems that enable Coinbase to effectively manage risk for its products and continue to build trust with its users.

Our Machine Learning system consists of high-quality labels, features, and models. We collect unbiased labels from fraud reversals, customer reports, holdout groups, and manual reviews. Then we engineer thousands of features ranging from traditional user localization, browser data, and authentication type to on-chain transaction history and knowledge graph. Afterward, our team leverages state-of-the-art fraud detection model architectures, including boosted trees, sequence models, and neural networks, to achieve higher precision and recall in detecting nefarious activity. Before production, we also conduct rigorous backtesting and A/B testing to ensure maximized customer protection with minimized false alarms. Throughout the workflow, a robust machine learning platform is in place to support batch and real-time feature stores, one-click training, high availability and low latency serving, as well as automatic data refresh.

Ultimately, our machine learning models are built to minimize the effect of bad actors on our exchange infrastructure. We understand that the world of cybersecurity is in constant flux, which is why we want you to know that we will continue to do our best to protect your accounts and digital assets. Through machine learning we are better able to provide the security measures necessary to best safeguard the entire community. 

We’re proud to lead the way in providing the best security protections available to all of our 103+ million users. Trust is built on dependable security — which is why we make protecting your account & your digital assets our number one priority. Learn more here.

Coinbase logo