Semi-Supervised Risk Model Training with Self-Generated Labels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fraud detection systems in e-commerce face challenges due to the scarcity of properly labeled data transactions, leading to missed fraudulent instances and inefficiencies in human review processes.
Innovation Solution
A semi-supervised machine learning method that utilizes a small set of labeled data transactions to train a supervised risk determination model, which then scores unlabeled transactions and iteratively expands the training set with newly labeled data, allowing for continuous improvement of fraud detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If human review process is used to label data transactions, then labeled data transactions are obtained for training, but the process is time-consuming and resource-intensive
Solution Approach 1:
The system performs self-labeling by using the trained risk determination model to automatically assign risk scores and labels to unlabeled data transactions. This eliminates the need for continuous human review to expand the training set, as the model serves itself by generating labeled data from its own predictions on unlabeled transactions.
Solution Approach 2:
The system performs preliminary labeling by using the initial labeled data to train a model that then pre-labels a large portion of unlabeled transactions. This preliminary action creates a expanded training set that can be used for further model training, reducing the need for time-consuming human review of all transactions.
2Measurement precision
If more labeled data transactions are used to train the risk determination model, then fraud detection accuracy is improved, but the cost and resources increase
Solution Approach 1:
The model expands its own training set by automatically labeling unlabeled transactions using its current risk assessment capabilities. This self-service approach allows the system to improve its training data quantity without proportionally increasing human review resources or costs.
Solution Approach 2:
The risk determination model acts as an intermediary between the initial small set of labeled transactions and the large volume of unlabeled transactions. It translates the limited labeled data into a expanded training set, enabling improved accuracy without directly proportionate resource investment.
3Loss of energy
If a supervised model is trained only on a small set of labeled transactions, then training resources are reduced, but the model misses fraudulent instances due to insufficient training data
Solution Approach 1:
The model improves its own reliability by automatically generating additional labeled training data from unlabeled transactions. This self-service mechanism allows the model to iteratively expand its training set and improve its fraud detection capabilities without requiring proportional increases in initial human review resources.
Solution Approach 2:
The system implements continuous model improvement by repeatedly using the risk determination model to label new unlabeled transactions, adding them to the training set, and retraining the model. This continuous cycle of labeling and retraining progressively enhances fraud detection reliability over time.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Training risk determination models based on a set of labeled data transactions. A first set of labeled data transactions that have been labeled during a review process is accessed. A first risk determination model is trained using the first set of labeled data transactions. A first risk score for data transactions of a set of unlabeled data transactions is determined using the first risk determination model. Data transactions in the set of unlabeled data transactions are newly labeled based on the first risk score. The newly labeled data transactions are added to a second set of labeled data transactions that include the first set of labeled data transactions. A second risk determination model is trained using at least the second set of labeled data transactions. A second risk score is determined for subsequently received data transactions and these data transactions are rejected or approved based on the second risk score.