Hierarchical Card Inactivity Prediction Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine-learning (ML) approaches face challenges in predicting card inactivity due to high data variability and scalability issues, particularly in the context of credit card transactions, where predicting inactivity is crucial for financial institutions but is hindered by the large volume of transactions and variability in user behavior over time.
Innovation Solution
The implementation of a hierarchical data modeling system that includes a first level classifier to determine whether an account will attrite and a second level classifier to predict when attrition will occur, using features such as transaction data and proper label creation, along with domain-specific features and data preprocessing to improve model performance and scalability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If machine-learning approaches are used to predict card inactivity, then insights can be provided on large-scale data for classification tasks, but accuracy deteriorates due to high variability between training data and validation data
Solution Approach 1:
The patent segments the classification task into two hierarchical levels: a first level classifier that determines whether an account will attrite, and a second level classifier that predicts when attrition will occur. This segmentation allows each classifier to focus on a specific aspect of the prediction problem, improving overall accuracy despite data variability. The first level classifier handles the binary decision of attrition occurrence, while the second level classifier handles the timing prediction, thereby addressing the accuracy issue mentioned in the contradiction.
2Measurement precision
If machine-learning approaches are used to predict card inactivity, then classification insights can be obtained, but scalability deteriorates due to billions of transactions and cards globally
Solution Approach 1:
By segmenting the classification into two levels, the system improves scalability. The first level classifier processes all accounts to identify those at risk of attrition, and only the second level classifier is applied to the subset of accounts predicted to attrite. This hierarchical segmentation reduces the computational burden on the second level classifier, enabling the system to scale to billions of transactions and cards while maintaining classification accuracy.
3Productivity
If traditional ML classification is applied to card activity data, then predictions can be made, but reliability deteriorates due to high variability in card activity over time
Solution Approach 1:
The hierarchical segmentation divides the prediction task into two reliable stages: first determining if attrition will occur, then predicting when it will occur. This segmentation improves reliability by ensuring that the second level classifier only makes timing predictions for accounts that have already been identified as attriting, reducing the impact of data variability on overall prediction reliability.
4Device complexity
If a single ML model is used for attrition prediction, then the system remains simple, but accuracy deteriorates due to the complexity of predicting both whether and when inactivity occurs
Solution Approach 1:
The patent applies segmentation by creating two specialized classifiers instead of one general model. The first level classifier specializes in determining attrition occurrence, while the second level classifier specializes in predicting attrition timing. This segmentation allows each model to be optimized for its specific task, improving overall prediction accuracy while keeping individual model structures relatively simple.
Data Source
AI summary
Systems and computer-implemented methods are described for modeling card inactivity. For example, hierarchical modeling may be used in which a first level classifier may be trained and validated to predict whether a card will be inactive. For cards predicted to become inactive by the first level classifier, a second level classifier may be trained and validated to predict when the card will become inactive. The first level classifier may include a binary classifier that generates two probabilities that respectively predict that the card will and will not become inactive. The second level classifier may include a multi-class classifier that generates a first probability that the card will become inactive at a first time period (such as one or more months in the future) and a second probability that the card will become inactive at a second time period. The multi-class classifier may generate other probabilities corresponding to other time periods.