Machine Learning Image Classification for Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems face challenges in efficiently analyzing large data sets to detect fraudulent behavior due to the complexity of defining patterns and high memory and computing resource consumption.
Innovation Solution
A system that utilizes a machine-learning image-classification model trained with depth chart images to identify patterns in transactional data, reducing dataset size by using 'depth of book' data and generating depth charts to detect behaviors of interest, such as market manipulation or fraud.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional rule-based systems are used to analyze large amounts of data to detect fraudulent behavior, then detection capability is provided, but memory and computing resources are significantly consumed
Solution Approach 1:
The patent extracts only the essential features needed for fraud detection from the complete transaction data, creating a reduced feature set that maintains detection capability while significantly reducing computational requirements. This is achieved by identifying and retaining only the most discriminative features for pattern recognition.
Solution Approach 2:
The patent transforms the data representation by changing parameters from raw transaction data to depth chart images with specific visual features. This parameter transformation enables the use of image classification algorithms that are more computationally efficient than traditional rule-based analysis of raw transaction data.
2Reliability
If conventional rule-based systems define complex patterns to detect fraudulent behavior, then detection accuracy can be improved, but the system becomes difficult and error-prone to define and maintain
Solution Approach 1:
The patent replaces manual rule-based pattern definition with an automated machine learning image classification system. The complex pattern recognition task is substituted from a manual mechanical process to an automated computational process that learns patterns from training data, eliminating the complexity of manual pattern definition while maintaining high detection accuracy.
3Reliability
If complete data sets are analyzed to ensure accurate fraud detection, then detection reliability is improved, but processing time increases
Solution Approach 1:
The patent extracts and processes only the most relevant features from complete transaction datasets, converting them to depth chart images that capture essential patterns. This extraction approach maintains detection reliability by preserving critical information while reducing the volume of data requiring processing, thereby decreasing processing time.
Solution Approach 2:
The patent performs preliminary transformation of transaction data into depth chart images before analysis, pre-processing the data into a format optimized for rapid image classification. This preliminary action enables faster processing while maintaining the integrity and reliability of fraud detection by preserving essential pattern information.
Data Source
AI summary
Systems and methods for identifying a pattern in data to detect a behavior of interest. The systems and methods receive a data stream representing a series of events occurring over a time interval. The systems and methods determine, for the interval, a depth indicating the amount of interest during the interval, where the interest represents an amount of a selected activity during that interval for a selected parameter. The systems and methods also generate a depth chart for the activities at different values of the selected parameter over a series of intervals, train a machine-learning image-classification model using depth chart images; identify the behavior of interest using predictions from the trained machine-learning model and target pattern; and provide an indication of the presence or absence of the behavior of interest. The parameter can represent a quantity and associated price of a commodity in a market.