Unexpected events such as production faults, system defects, and suspicious activity can be difficult to anticipate. By analyzing historical patterns across multiple entities, machine learning systems may identify warning signs that help predict anomalies before they occur.
Mastercard proposes a machine learning system that analyzes historical anomalies, generates anomaly scores, and uses those scores to predict future anomalous events.

How it works
The system organizes each entity’s features into sets corresponding to different time steps, allowing it to examine changes in the data over time. It then counts how many entities experienced anomalies at specific points in the past. When this count meets a predefined threshold, the system assigns an anomalous label to each entity in the set.
The machine learning model uses these feature sets and labels to calculate an anomaly score for the group. The score is then compared against a separate prediction threshold to determine whether the system should predict an upcoming anomalous event.
By separating historical anomaly labeling from future-event prediction, the system uses two thresholds to distinguish patterns in past data from those that warrant a prediction.
Why it matters
Predicting anomalous events before they occur could help organizations identify potential risks earlier. In financial services, machine learning techniques can support fraud detection by analyzing large volumes of transaction data to identify suspicious patterns and help firms respond more quickly.
The patent application aligns with Mastercard’s ongoing use of artificial intelligence and machine learning to detect unusual activity and assess transaction risks. The company’s existing fraud-prevention solutions use data-driven analysis to identify potentially fraudulent transactions, while this proposed method focuses on analyzing historical anomalies to predict future events. Although the application does not establish that the method is already integrated into Mastercard’s products, it highlights the potential for machine learning to support earlier risk detection and more proactive fraud prevention.
U.S. Pat. App. Pub. No. 2026/0312969, titled “Methods and Systems for Predicting Upcoming Anomalous Events” was filed on April 04, 2025, and was published on October 08, 2026. The listed inventors are Alekhya Bhatraju, Aditi Rani, Debasmita Das, Nishant Pant, and Yatin Katyal.
