Self-service terminal fraud detection using multi-modal sensor fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Self-service terminals, such as ATMs, are vulnerable to fraudulent activities like card trapping and skimming, making it difficult to differentiate between legitimate and fraudulent behavior, and existing anti-fraud measures are inadequate in detecting such threats effectively.
Innovation Solution
A method using a classifier with multiple input zones, each receiving data of different modalities, combines weighted sets from various sensors using a dedicated weighting function to produce a composite set, which is then used to derive class values for fraud detection, employing kernels and regression techniques within a Bayesian hierarchical model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple sensors and data modalities are used to detect fraud, then the detection capability is improved, but the device complexity increases
Solution Approach 1:
The patent combines multiple sensors (video camera, audio sensor, proximity sensor, card reader) and their respective data modalities into a unified classifier system. The classifier integrates these diverse data sources through a common processing architecture, merging separate detection capabilities into a coordinated anti-fraud system that improves overall detection accuracy while managing complexity through integration.
2Measurement precision
If a classifier with multiple input zones and weighting functions is implemented, then the classification accuracy is improved, but the computational complexity increases
Solution Approach 1:
The classifier is divided into multiple input zones, each dedicated to processing specific sensor modalities (video, audio, proximity, card reader data). Each zone applies appropriate weighting functions tailored to its specific data type. This segmentation allows complex computational tasks to be distributed across specialized sub-units, improving classification accuracy while managing computational complexity through modular organization.
3Measurement precision
If Bayesian hierarchical models with kernels are used, then the fraud prediction accuracy is improved, but the processing time increases
Solution Approach 1:
The system pre-computes and stores kernel functions and their parameters during a training phase, before actual fraud detection is needed. The Bayesian hierarchical model parameters are预先 estimated and stored. During real-time operation, the system only needs to evaluate these pre-computed models against new sensor data, significantly reducing processing time while maintaining high prediction accuracy.
Data Source
AI summary
A method of detecting attempted fraud at a self-service terminal. The method involves providing a classifier having a plurality of input zones, each input zone being adapted to receive data of a different modality. The classifier may be a statistical model. The method also involves mapping inputs from a plurality of sensors in the self-service terminal to respective input zones of the classifier. For each input zone, the classifier independently operates on inputs within that zone to create a weighted set; combines the weighted sets using a dedicated weighting function for each weighted set to produce a composite set; and derives a plurality of class values from the composite set to create a classification result. The classification result can be used to predict the probability of abnormal operation, which may be indicative of fraud.