DataRobot has been nominated for the 2026 World AI Awards in the Anomaly Detection category, recognising its work in using machine learning to identify unusual patterns buried inside large and often messy datasets.
It is a problem that sounds straightforward until the data arrives. Fraudulent transactions, equipment faults, cybersecurity incidents and other abnormal events are usually rare. The overwhelming majority of records describe normal behaviour. Worse, organisations may not have enough correctly labelled examples of previous incidents to train a conventional predictive model.
That is precisely where DataRobot’s anomaly detection technology is designed to operate.
The World AI Awards recognises organisations, individuals, products and technologies contributing to the development and practical application of artificial intelligence across industries.
For DataRobot, anomaly detection sits inside a broader enterprise AI platform spanning predictive AI, generative AI, agentic systems, deployment, governance and observability.
Finding the unusual when nobody has labelled it
DataRobot supports unsupervised anomaly detection, meaning organisations can build models without specifying a conventional target variable.
Instead of learning from a dataset where every suspicious event has already been identified, the system examines patterns in the available data and assigns anomaly scores to observations that differ from expected behaviour. DataRobot supports methods including Isolation Forest, One-Class Support Vector Machine, Local Outlier Factor and Double Mean Absolute Deviation.
That matters when abnormal events are both uncommon and difficult to label.
DataRobot points to network and cybersecurity events, insurance fraud and credit-card fraud as examples. Its documentation also demonstrates an anti-money-laundering workflow in which historical transaction data is analysed without using the existing fraud labels to train the anomaly model.
The idea is not to declare every unusual transaction fraudulent. An anomaly is a signal for investigation, not proof of wrongdoing.
An anomaly score is only useful if someone can investigate it
Finding an unusual row in a dataset is one thing. Understanding why the model considered it unusual is harder.
DataRobot generates an anomaly score for observations, allowing users to rank records according to how strongly they deviate from learned patterns. Its platform can then provide model insights such as Feature Impact and Prediction Explanations to help investigate the factors contributing to an outlier designation.
For DataRobot-built anomaly detection blueprints, Feature Impact is calculated using SHAP, an established approach for explaining how individual features influence model outputs.
There is another complication: an unsupervised model has no ground-truth target against which conventional predictive accuracy can automatically be measured.
DataRobot addresses that with Synthetic AUC, a metric that generates synthetic examples representing more normal and more anomalous observations and uses them to help compare candidate models. DataRobot itself describes Synthetic AUC as an approximation, an important limitation when interpreting the metric.
Watching machines for the signal before failure
Anomaly detection becomes particularly interesting when the data is arriving continuously.
DataRobot supports time-series anomaly detection, where the system looks for abnormal behaviour as new observations arrive rather than simply analysing a static historical dataset.
Consider an industrial pump.
Sensors may continuously report pressure and other operating measurements. A reading that suddenly moves outside the machine’s normal behaviour could indicate a developing problem. DataRobot uses this scenario in its documentation to illustrate how an anomaly model could flag abnormal pressure so maintenance teams can investigate before an eventual pump failure.
That is predictive maintenance in practical terms: not predicting that every flagged machine will fail, but identifying behaviour unusual enough to justify attention.
The same underlying concept can travel across industries. A strange financial transaction, unexpected network activity and an abnormal sensor reading are very different events, but all present a similar machine-learning problem — finding the few observations that do not look like everything around them.
Anomaly detection moves into AI monitoring itself
DataRobot is also applying anomaly detection to the systems running AI.
Its current AI Observability capabilities monitor agents and AI applications for behaviour, output quality, data inputs, system interactions and performance. The platform includes anomaly detection with configurable alert thresholds alongside monitoring for performance degradation and changes in incoming data.
That creates an interesting loop.
AI can be used to identify anomalies in business data, while observability technology watches the AI systems themselves for unexpected changes.
DataRobot says its monitoring can also detect model-performance drift and data-quality problems, with alerts and diagnostic tools intended to help teams investigate issues before they spread further through production systems.
Graham Cooke, President of the World AI Awards, said:
“DataRobot’s nomination in Anomaly Detection highlights an important part of practical artificial intelligence: finding the signals that people may struggle to identify inside enormous volumes of normal activity.
“Whether the application involves financial transactions, cybersecurity, industrial equipment or the performance of AI systems themselves, anomaly detection can give organisations an earlier indication that something deserves attention. We congratulate DataRobot on its 2026 World AI Awards nomination and look forward to following the continued development of its technology.”
DataRobot joins organisations, researchers, entrepreneurs and technology developers being recognised through the 2026 World AI Awards.
The programme recognises organisations, individuals and technologies contributing to the development and application of artificial intelligence across industries. Anomaly detection represents a particularly practical corner of that landscape because the technology is often concerned not with generating something new, but with noticing what does not belong.
DataRobot’s approach reflects that challenge. Its anomaly detection capabilities can work with unlabeled and partially labelled information, extend into time-series environments and provide explanations intended to help people investigate why observations have been flagged. The platform also carries the concept into AI observability, where abnormal behaviour inside production AI systems can itself become something worth investigating.
Learn more about DataRobot and its enterprise AI technology at https://www.datarobot.com/.
Discover the World AI Awards 2026, explore the nominees and learn more about the awards at https://www.worldawards.ai/.
