IBM Safer Payments introduces an AI-powered Feature Generator that replicates human domain expertise to automatically create, evaluate and optimize behavioral features.
With version 6.9, IBM Safer Payments introduces an AI-powered Feature Generator that replicates human domain expertise to automatically create, evaluate and optimize behavioral features through an iterative, evolutionary process enabling faster, more accurate and more explainable fraud-detection models.
In payment fraud prevention, the quality of a model depends on the quality of its features, the behavioral profiles that capture how entities act over time. Traditional feature engineering methods such as combining, aggregating, or transforming existing data are useful but limited; they don’t create the behavioral profiling features fraud experts rely on to detect complex and evolving attack patterns.
Creating these behavioral features has always been the most time-consuming and expertise-dependent step in model development.
Automating feature generation increases efficiency and scalability in machine-learning workflows, reducing manual effort and potential for human error while enabling the discovery of novel and effective features that traditional manual methods often miss.
This new capability goes beyond process automation; it applies a proprietary AI/ML algorithm that replicates human domain expertise. The algorithm automatically generates and evaluates meaningful behavioral features (counters or profiles) that directly impact fraud-detection model performance. By using AI not only for model training but also for the complex step of feature creation, model quality, explainability and predictive accuracy are significantly enhanced.
Traditional feature creation is tedious, time-consuming and highly dependent on a data professional’s domain knowledge and intuition. Manual feature selection often lacks standardization, so different analysts or even the same analyst over time may select different sets of features, resulting in inconsistent models and outcomes.
With AI-powered automation, Safer Payments delivers a standardized, reproducible and unbiased approach to feature generation. The process reduces variability and ensures dependable results and empowers fraud teams to produce high-performing models faster transforming model development from a manual art into an intelligent, automated science.
The new capability is conveniently embedded in IBM Safer Payments’ ‘Model Factory’, the collection of tools allowing users to rapidly train, test, and deploy custom machine-learning models and rules to combat emerging threats.
Using an intuitive interface, the new capability allows users to:
Various data analytic techniques are used to generate a population of candidate features by analyzing the training dataset. In an iterative evolutionary process, the algorithm simulates, evaluates, and scores each candidate to determine its predictive importance and overall impact on model performance. In each iteration, the highest-performing features are retained while lower-performing candidates are replaced with new variations using selection, crossover, and mutation. Once the process completes, the evaluated features are presented with detailed scoring and statistics that illustrate their relevance, detection lift, and contribution to false-positive reduction.
By embedding AI-powered automation directly into the model-building process, IBM Safer Payments enables fraud teams to react instantly to new attack patterns, improving detection accuracy, reducing false positives and cutting operational effort.
This capability doesn’t just speed up model development, it completes the end-to-end automation of fraud-model creation, connecting data ingestion, feature generation and model training into one seamless workflow within the Model Factory. Fraud analysts can now build, test and deploy sophisticated, explainable models without relying on external data-science resources or vendor intervention.
The result is a more agile, transparent, and self-sufficient fraud-prevention ecosystem empowering institutions to respond faster to emerging threats, ensure regulatory confidence and continuously improve model quality at scale.
AI Assisted Feature Generation is included in IBM Safer Payments v6.9, available now for deployment.