Adaptive Feature Extraction for Abnormal Financial Transaction Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for detecting abnormal financial transactions are ineffective in identifying new patterns, leading to a high false negative rate for unknown abnormal transactions.
Innovation Solution
An electronic apparatus and method that preprocesses payment data, extracts features based on adaptive sampling rates, and uses a machine learning algorithm to determine abnormal transactions, ensuring high detection rates for both known and new patterns through unsupervised learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If preset rules based on past transaction data are used to detect abnormal transactions, then detection rate for known abnormal patterns is high, but detection capability for new abnormal patterns is poor
Solution Approach 1:
The patent applies dynamics by transitioning from static preset rules to dynamic machine learning models that continuously adapt to new transaction patterns. The system retrains models with accumulated transaction data, enabling the detection mechanism to evolve and recognize both known and emerging abnormal patterns effectively.
Solution Approach 2:
The patent changes the fundamental parameter of detection from fixed rule-based thresholds to adaptive machine learning parameters. By using algorithms that learn from data and automatically adjust their parameters based on transaction patterns, the system achieves both high precision for known patterns and adaptability for new patterns.
2Adaptability or versatility
If machine learning algorithms are used to detect new abnormal transaction patterns, then adaptability to new patterns improves, but detection speed and real-time processing capability may deteriorate
Solution Approach 1:
The patent applies preliminary action by pre-training machine learning models with historical transaction data before actual detection occurs. The models are prepared in advance with learned patterns, enabling them to rapidly classify new transactions without requiring complex real-time computation, thus maintaining high processing speed while detecting new abnormal patterns.
Solution Approach 2:
The patent segments the detection process into offline model training phase and online real-time classification phase. The computationally intensive training occurs separately from time-critical transaction processing, allowing the system to use sophisticated machine learning algorithms during training while maintaining fast response times during actual transaction detection.
3Measurement precision
If comprehensive feature extraction from all payment data items is performed, then detection accuracy improves, but processing time and computational complexity increase
Solution Approach 1:
The patent extracts only the most relevant features from payment data for machine learning analysis, rather than processing all available data items. By identifying and extracting key features that have the highest predictive value for abnormal transaction detection, the system maintains high detection accuracy while significantly reducing processing time and computational requirements.
Data Source
AI summary
Provided are a method of detecting abnormal financial transactions and an apparatus thereof. The apparatus includes: a memory in which an abnormal transaction detection program is stored; and a processor configured to execute the program. Upon execution of the program, the processor performs a data preprocessing operation to acquired payment data, extracts at least one feature adaptively determined in advance from results of the preprocessing operation, and uses the extracted feature to determine whether the payment data correspond to an abnormal transaction through a machine learning algorithm adaptively determined in advance.