SAS® Adaptive Learning and Intelligent Agent System Features
Anomaly detection
- Combines supervised and unsupervised machine learning algorithms to detect fraud and other rare events.
- Supports exporting/publishing of models to SAS Model Manager where other SAS tools can access the model.
- Includes APIs that give you the flexibility to use other systems for record scoring, alert generation, classification and model retraining.
- Presents charted anomaly detection results to help set alert thresholds and control volume of alerts generated.
- Highlights top variables contributing to the anomalies through charts in the data set.
- Provides an unsupervised scorecard to help interpret why a row was identified as anomalous.
- Alerts on a percentage of records from below the threshold (below the line), reducing the potential for introducing bias into the model over time.
- Enhances job monitoring – including improvements to job status, error handling and logging – which increases the accuracy of job progress status reporting.
- Improves efficiency by duplicating models based on other model versions, reducing time to adjust variations of models for best results.
Transparency
- Includes a white box model and record scorecard to take the mystery out of machine learning.
- Delivers narrative, text and visuals for describing and interpreting models.
- Allows analysts and investigators to provide transparency into decisioning and explain why someone was identified as fraudulent.
Adaptive learning
- Creates an adaptive learning system that continually improves its ability to detect fraud and react rapidly to the emergence of new fraud patterns via integration with SAS Visual Investigator.
- Feeds triage results (e.g., alert productivity ratings) from SAS Visual Investigator back into the training data for training new models.
- Supports the entire analytics life cycle – from data, to discovery, to deployment.
Easy-to-use visual interface
- Creates models for detecting rare events by selecting:
- Training data tables that include records classified for fraud.
- Target column containing fraud classifications.
- Values that indicate fraud.
- Model complexity.
- Variables to exclude or include.
- Evaluates model effectiveness and selects appropriate alerting threshold.
- Provides charts describing model accuracy. Compares values in the target column to the model’s classification of each record and charts the balance of false positives and negatives for each threshold.
- Automates best threshold selection.
- Lets you override the threshold with a threshold of your own.
- Enables you to select a threshold and see the impact on false positives and negatives.
- Supports multiple versions of a model for evaluation prior to deployment.
- Deploy models at the click of a button, automatically creates a set of jobs for generating scores and alerts.
Automated integration
- Automatically integrates with SAS Visual Investigator for scoring, alerting, triaging, refreshing training data and retraining models.