Fuzzy Tagging for Transaction Clustering and Fraud Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current transaction tagging methods lack granularity, often categorizing transactions as either fraudulent or non-fraudulent, leading to overprediction of fraudulent transactions and a lack of certainty in prediction accuracy.
Innovation Solution
A computer-implemented method for fuzzy tagging that forms clusters of transactions based on initial tag values, determines centroids for these clusters, and calculates fuzzy tag values using the distance of transactions from these centroids, allowing for a more nuanced prediction of transaction certainty through the use of a fuzzy membership matrix.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If binary tag values (0 or 1) are used to categorize transactions, then the tagging process is simple and computationally efficient, but the prediction lacks granularity and overpredicts fraudulent transactions
Solution Approach 1:
The patent transforms the discrete binary tag parameters (0 or 1) into continuous fuzzy tag values ranging from 0 to 1. This parameter change allows transactions to be represented with degrees of fraud likelihood rather than absolute categories, thereby improving prediction accuracy while maintaining computational efficiency through standardized mathematical operations.
Solution Approach 2:
The patent introduces a new dimensional aspect to transaction tagging by adding the fuzzy membership value dimension. Instead of solely relying on binary classification, the system now operates in a continuous value space, enabling nuanced representation of transaction risk and reducing overprediction of fraudulent transactions.
2Measurement precision
If fuzzy tag values are calculated using distance from centroids, then prediction certainty is improved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary clustering of transactions into fraud and non-fraud groups before calculating fuzzy tags. By pre-establishing centroids for these clusters, the system reduces the computational burden during real-time tagging, as each transaction only needs to calculate distances to pre-computed centroids rather than performing complex classification from scratch.
Solution Approach 2:
The patent uses centroid representations as simplified copies or proxies for entire clusters of transactions. Instead of comparing each transaction against all historical data points, the system uses the centroid (average characteristics) of each cluster as a representative copy, significantly reducing computational complexity while maintaining prediction accuracy.
3Loss of information
If transactions are clustered into multiple groups, then granularity of prediction is improved, but the complexity of determining centroids and distances increases
Solution Approach 1:
The patent segments the transaction data into distinct clusters based on fraud likelihood, creating multiple groups rather than a single homogeneous set. This segmentation preserves information granularity by maintaining the distinct characteristics of different transaction types while organizing them into manageable clusters with representative centroids.
Solution Approach 2:
The patent merges multiple transactions with similar characteristics into unified clusters, combining their features to form representative centroids. This merging process reduces the complexity of handling individual transactions by grouping them, while still preserving the nuanced information through the cluster's central characteristics.
Data Source
AI summary
A computer-implemented method of tagging a transaction that includes tagging a transaction with one of a first tag value or a second tag value, forming a set of clusters associated with the tagged transactions having a first value, and forming a second set of clusters associated with the tagged transactions having a second value. The computer implemented method also includes determining a fuzzy tag value based on a relationship between the transaction and one of the centroids of the clusters having a first tag value, and one of the centroids of the clusters having second value. The method also includes replacing the first tag value or the second tag value with the fuzzy tag value.