Smoothing Sparse Multi-Dimensional Risk Tables for Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for detecting fraudulent transactions face challenges in constructing effective models due to the sparseness of data in multi-dimensional risk tables, leading to many blank values and inaccurate risk assessments.
Innovation Solution
The method involves approximating initial risk values for empty cells in multi-dimensional risk tables using weighted sums of non-empty cells and applying adjustment values to improve accuracy, employing incremental factorization-based smoothing to refine predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multi-dimensional risk tables are used to capture interactions among categorical variables, then measurement precision of risk assessment is improved, but the sparseness of data causes many blank values leading to unreliable risk estimates
Solution Approach 1:
The patent introduces smoothing techniques as an intermediary process between the sparse multi-dimensional risk table data and the final risk estimates. By applying smoothing algorithms, the system interpolates missing values in the risk table using information from related cells, thereby providing reliable risk estimates even when direct observations are sparse or absent.
Solution Approach 2:
The patent transforms the sparse risk table data by applying parameter changes through smoothing operations. This involves modifying the raw risk values by incorporating information from neighboring or related cells in the risk table, effectively changing the parameters (risk estimates) to reflect both observed data and inferred patterns, thus improving reliability without sacrificing the multi-dimensional interaction capture.
2Measurement precision
If multi-dimensional risk tables are constructed to capture variable interactions, then measurement precision is improved, but device complexity increases due to the large number of cells and data requirements
Solution Approach 1:
The patent segments the complex multi-dimensional risk table into manageable components by applying smoothing operations that process subsets of cells independently or semi-independently. This segmentation allows the system to handle the complexity of multi-dimensional interactions without requiring the entire risk table to be fully populated, as smoothing can be applied incrementally to fill gaps using local information.
Solution Approach 2:
The patent applies partial action by using smoothing techniques that do not require complete data across all dimensions of the risk table. Instead of demanding full population of every cell, the smoothing process performs partial inference using available data from related cells, thereby achieving reliable risk estimates without the excessive complexity of requiring complete multi-dimensional data coverage.
3Measurement precision
If historical models are constructed from large numbers of transactions, then measurement precision of fraud detection is improved, but the difficulty of deploying these models in real-time production environments increases
Solution Approach 1:
The patent performs preliminary action by pre-computing smoothed risk estimates and storing them in the multi-dimensional risk table during the model training phase. This preliminary smoothing of historical data creates a pre-processed structure that can be directly applied to real-time transactions without requiring complex computations during deployment, thereby maintaining high accuracy while improving ease of manufacture and deployment.
Solution Approach 2:
The patent uses copying by creating a structured representation of historical risk patterns in the form of a multi-dimensional risk table with pre-applied smoothing. This copied structure serves as a lookup table or reference model that can be efficiently applied to new transactions, replicating the complex analysis performed on historical data without re-executing it, thus enabling easy deployment in real-time environments while maintaining measurement precision.
Data Source
AI summary
A system for classifying a transaction as fraudulent includes a training component and a scoring component. The training component acts on historical data and also includes a multi-dimensional risk table component comprising one or more multidimensional risk tables each of which approximates an initial risk value for a substantially empty cell in a risk table based upon risk values in cells related to the substantially empty cell. The scoring component produces a score, based in part, on the risk tables associated with groupings of variables having values determined by the training component. The scoring component includes a statistical model that produces an output and wherein the transaction is classified as fraudulent when the output is above a selected threshold value.