Cryptocurrency MAT Detection Using Balanced Feature-Based Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting multi-party anonymization transactions (MATs) in cryptocurrency transactions are inadequate, particularly for advanced techniques, as they often rely on rule-based and heuristic approaches that fail to address the complexity of these transactions.
Innovation Solution
A computerized method involving the use of a trained entry classifier model that optimizes training data features and subsets, employing machine learning techniques to accurately classify transactions as either MATs or non-MATs, using labeled training data from public and private sources, and balancing the data sets to enhance model efficiency and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If rule-based and heuristic approaches are used for detecting MATs, then the detection method is simple to implement, but the detection accuracy is insufficient for advanced MAT techniques
Solution Approach 1:
The patent replaces rule-based and heuristic detection methods (mechanical/systematic approaches) with machine learning models that automatically learn detection patterns from data. The entry classifier model and transaction classifier model are trained on labeled transaction data to automatically identify MAT transactions, substituting manual rule creation with automated learning systems that achieve higher accuracy.
Solution Approach 2:
The patent transforms the detection approach by changing from fixed rules to dynamic parameters learned from data. The machine learning models adjust their internal parameters (weights and biases) based on training data, enabling them to adapt to various MAT techniques. This allows the system to detect both simple and advanced MAT transactions by learning their characteristic patterns.
2Measurement precision
If machine learning models are trained with all available training data features, then the classification accuracy improves, but the processing resources and time required increase significantly
Solution Approach 1:
The patent extracts and selects only the most relevant features for training the machine learning models. The entry classifier model identifies which data features are most important for detecting MAT transactions, and the transaction classifier model similarly selects relevant features from the extracted entry features. This feature selection process removes redundant and less important features, reducing processing requirements while maintaining detection accuracy.
Solution Approach 2:
The patent segments the classification process into two distinct stages: first, the entry classifier model determines whether individual transaction entries are MAT-related; second, the transaction classifier model analyzes groups of entries to classify the overall transaction. This segmentation allows each model to focus on specific aspects, improving efficiency and reducing the computational burden compared to a single monolithic classifier.
3Productivity
If the training data set is balanced with equal representation of MAT and non-MAT classes, then the model training becomes more efficient, but the natural distribution of transactions is altered
Solution Approach 1:
The patent applies oversampling to the minority class (MAT transactions) and/or undersampling to the majority class (non-MAT transactions) to create a balanced training dataset. This parameter change in data distribution allows the machine learning models to learn from equal representation of both classes, improving training efficiency and preventing bias toward the majority class, while the models generalize well to the natural distribution through proper evaluation on unbalanced test data.
Data Source
AI summary
A computerized method trains an entry classifier model and uses the entry classifier model to identify multi-party anonymization transaction (MAT) entries. Labeled training data is obtained from a training data source and standard data features are identified therein. Engineered data features are generated using the identified standard data features. Training data features are selected from the standard data features and the engineered data features. A balanced training data subset is generated using the obtained labeled training data and an entry classifier model is trained to classify data entries as being in a MAT class based on the selected training data features using the balanced training data subset. The trained entry classifier model is used to classify an input data entry as being in the MAT class.