Training Dataset Generation for Imbalanced Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning models for anomaly detection, particularly in fraud detection systems, face challenges due to extreme class imbalance in datasets, leading to inaccurate classifications with high rates of false positives and false negatives, as they are trained on imbalanced datasets with rare fraudulent events outnumbered by numerous non-fraudulent transactions.
Innovation Solution
A computerized system generates a reduced-size superior labeled training dataset using a processor that retrieves instances from minority and majority labeled class databases, applies clustering and learner models with differentiability and indicativeness estimators to select most distant instances, and iteratively refines the dataset until a preconfigured condition is met, providing a high-accuracy training dataset for machine learning classification models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning models are trained on extremely imbalanced datasets with rare fraudulent events, then the model can detect rare anomalies, but the classification accuracy deteriorates with high rates of false positives and false negatives
Solution Approach 1:
The patent extracts and isolates the minority class instances (fraudulent events) from the imbalanced dataset to create a separate focused training subset. By extracting only the most informative minority instances and pairing them with carefully selected majority instances, the system overcomes the limitation of training on extremely imbalanced data while maintaining the ability to detect rare anomalies.
Solution Approach 2:
The patent applies local quality by creating region-specific training subsets tailored to different operational contexts or fraud types. Instead of using a single uniform training approach for the entire imbalanced dataset, the system generates specialized training sets with locally optimized instance selections that address specific detection challenges in different regions or domains.
2Productivity
If a reduced size training dataset is generated to improve training efficiency, then the training time and computational resources are reduced, but the model performance may deteriorate due to insufficient training data
Solution Approach 1:
The patent changes the parameters of data selection by introducing multiple criteria (frequency thresholds, recency weights, diversity metrics) to transform the original imbalanced dataset into a optimized training subset. By adjusting these selection parameters, the system generates training sets that are smaller than the original dataset but contain higher quality instances, thus improving training efficiency without sacrificing model performance.
Solution Approach 2:
The patent performs preliminary actions by pre-processing and pre-selecting training instances before the actual model training begins. The system identifies and prepares the most informative minority instances and their corresponding majority instances in advance, creating ready-to-use training subsets that eliminate the need for extensive computation during the training phase while ensuring high model performance.
3Loss of information
If all instances in the majority class are used for training, then the model learns from comprehensive data, but the training dataset size becomes excessively large increasing computational complexity
Solution Approach 1:
The patent extracts a representative subset of majority class instances that are most relevant for training. By applying selection criteria such as frequency thresholds, recency weights, and diversity metrics, the system extracts only the necessary majority instances paired with minority instances, eliminating redundant data while preserving essential information patterns.
Solution Approach 2:
The patent applies partial action by selecting only the necessary portion of majority class instances required for effective training. Instead of using all available majority instances, the system identifies and uses a partial subset that provides sufficient information for learning while avoiding the computational burden of processing excessive redundant data.
Data Source
AI summary
A computerized-system and method for generating a reduced-size superior labeled training-dataset for a high-accuracy machine-learning-classification model for extreme class imbalance by: (a) retrieving minority and majority class instances to mark them as related to an initial dataset; (b) retrieving a sample of majority instances; (c) selecting an instance to operate a clustering classification model on it and the instances marked as related to the initial dataset to yield clusters; (d) operating a learner model to: (i) measure each instance in the yielded clusters according to a differentiability and an indicativeness estimators; (ii) mark measured instances as related to an intermediate training dataset according to the differentiability and the indicativeness estimators; (e) repeating until a preconfigured condition is met; (f) applying a variation estimator on all marked instances as related to an intermediate training dataset to select most distant instances; and (g) marking the instances as related to a superior training-dataset.