Bank Dispute Scoring via Predictive Model and Shapley Values
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Bank dispute cases often require manual analysis by case managers, which is time-consuming and inefficient, as they need to individually assess evidence for each disputed transaction.
Innovation Solution
A computing device is used to analyze bank dispute cases by assigning values from transaction details to data features, which are then fed into a predictive model to generate approval or denial scores. These scores are accompanied by importance metrics, such as Shapley values, to provide explanations and facilitate case disposition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual analysis by case managers is used, then accuracy in assessing evidence can be maintained, but analysis time and efficiency deteriorate
Solution Approach 1:
The patent introduces a predictive model as an intermediary between the raw transaction data and the case manager's decision-making process. The model processes transaction details, account information, and dispute evidence to generate probability scores that assist case managers, thereby reducing their analytical burden while maintaining decision accuracy.
Solution Approach 2:
The patent replaces the mechanical manual analysis process with an automated predictive modeling system. The system uses machine learning models to process and evaluate evidence, substituting human manual assessment with algorithmic analysis that operates faster while maintaining or improving accuracy through consistent application of evaluation criteria.
2Reliability
If manual analysis by case managers is used, then nuanced judgment can be applied, but productivity and efficiency deteriorate
Solution Approach 1:
The predictive model serves as an intermediary that handles routine evaluation tasks, allowing case managers to focus on nuanced judgment for complex cases. The model provides probability scores and risk assessments that support rather than replace human expertise, maintaining nuanced judgment while improving overall productivity through automation of standard evaluations.
Solution Approach 2:
The system applies partial automation where the predictive model handles the quantitative assessment of evidence, while case managers retain responsibility for qualitative nuanced judgment. This partial action approach allows the system to improve productivity in measurable aspects while preserving human expertise for aspects requiring nuanced understanding.
3Measurement precision
If comprehensive transaction details are analyzed, then accuracy improves, but system complexity and data processing requirements worsen
Solution Approach 1:
The patent extracts and isolates the most critical features from comprehensive transaction details for input into the predictive model. By identifying and extracting key predictive features from large volumes of transaction data, the system maintains assessment accuracy while reducing the complexity of data processing requirements.
Solution Approach 2:
The system segments the comprehensive transaction analysis into distinct components: data extraction, feature selection, predictive modeling, and result interpretation. This segmentation allows each component to handle specific aspects of the analysis, reducing overall system complexity while maintaining the ability to process comprehensive transaction details for accurate assessment.
Data Source
AI summary
A computing device receives a request to analyze a bank dispute case comprising one or more disputed transactions. For at least a first disputed transaction, for each of a plurality of data features, the computing device determines, based on details of the respective disputed transaction, a data feature value for the respective data feature. Based on the respective data feature value of each respective value, the computing device determines a denial probability and a approval probability for the respective disputed transaction. In response to receiving an indication of user input selecting the first disputed transaction, the computing device generates a graphical user interface comprising at least an indication of the denial probability for the first disputed transaction, an indication of the approval probability for the first disputed transaction, and at least one indication of an importance metric for a first data feature of the plurality of data features.