Explainable Anomaly Detection via Expectation Surfaces
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current AI algorithms used in Anti-Money Laundering (AML) regimes provide predictions/detections of suspicious activities but fail to offer sufficient explanations, leading to skepticism among investigators and regulators, and a lack of trust in AI systems.
Innovation Solution
The development of a computer-implemented method that generates explanations for AI algorithm outputs by using an Expectation Surface, which provides an expected value range for features contributing to anomaly detection, thereby offering insights into why specific features led to the detection of suspicious activities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional AI algorithms are used for anomaly detection in AML, then detection accuracy is improved, but explainability deteriorates leading to skepticism and lack of trust
Solution Approach 1:
The patent introduces expectation surfaces as an intermediary component that bridges the gap between the AI anomaly detection model and human investigators. The expectation surface computes expected value ranges for input features and generates natural language explanations that mediate between the black-box model output and human understanding, thereby maintaining detection accuracy while improving explainability and trust.
2Reliability
If AI systems provide detailed explanations for anomaly detections, then trust and understanding are improved, but system complexity increases
Solution Approach 1:
The patent segments the explanation generation process into distinct modular components: (1) expectation surface computation module that calculates expected value ranges, (2) feature importance analysis module that identifies contributing factors, and (3) natural language generation module that formulates explanations. This segmentation allows each component to be developed and optimized independently, managing overall system complexity while providing comprehensive explanations.
3Productivity
If conventional AI algorithms are used in AML, then false positives are reduced, but investigators still lack understanding of detection drivers
Solution Approach 1:
The patent implements a feedback mechanism where the expectation surface continuously provides explanatory information back to investigators about the drivers behind anomaly detections. By analyzing the difference between actual feature values and expected value ranges, the system generates feedback explanations that highlight which features contributed most to the anomaly score, enabling investigators to quickly understand detection rationale without manual analysis.
Data Source
AI summary
The platforms, systems and methods provided herein may provide explanations for AI algorithm outputs to facilitate efficiency and trust for a user. More specifically, the platforms, systems and methods provided herein may provide anomaly detection using explainable machine learning algorithms. Provided here is a computer-implemented method for providing explanations for AI algorithm outputs, comprising: (a) receiving transaction log data; (b) identifying anomalous transactions based at least in part on the transaction log data; (c) generating an expectation surface for one or more anomalous transactions; and (d) generating explanations for the anomalous transactions based at least in part on the expectation surface.