Machine Learning Survey Prediction for Financial Assessment and Bias Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Financial assessment surveys face challenges such as low participation rates due to time and effort requirements, sampling bias, and inability to establish causal relationships between responses over time, making interventions unreliable and untimely.
Innovation Solution
A computing system uses a machine learning model with a neural network performing unsupervised learning to predict survey responses based on personal data, generating assessment scores or responses for users without requiring them to complete surveys, and takes actions such as sending communications or changing account settings based on these predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If financial health assessment surveys are conducted to evaluate respondent competency, then assessment accuracy is improved, but respondent participation rate deteriorates due to time and effort requirements
Solution Approach 1:
The patent creates a virtual copy of the survey response process by training a machine learning model on actual survey responses. The model then generates predicted responses for users without requiring them to complete the actual survey, thereby obtaining assessment data without the time and effort burden that reduces participation rates.
Solution Approach 2:
The patent performs preliminary action by collecting and analyzing survey responses in advance to train the machine learning model. This pre-training phase enables the system to generate accurate predictions later without requiring users to complete time-consuming surveys at the moment of assessment.
2Reliability
If follow-up surveys are administered to track changes in respondent well-being, then longitudinal assessment capability is improved, but sampling bias worsens due to reduced respondent participation
Solution Approach 1:
The patent uses the machine learning model to generate copies of survey responses for follow-up assessments. Since the model was trained on diverse respondent data, it can predict responses for users who would otherwise be unlikely to participate in follow-up surveys, thereby reducing sampling bias while maintaining longitudinal assessment capability.
Solution Approach 2:
The system performs self-service by using the trained model to generate predicted responses automatically without requiring user intervention. This enables continuous longitudinal tracking of user well-being changes without the participation barriers that cause sampling bias in traditional follow-up surveys.
3Ease of manufacture
If traditional survey methodology is used to establish causal relationships, then data collection simplicity is improved, but causal inference capability deteriorates due to insufficient data breadth and depth
Solution Approach 1:
The patent merges multiple data sources including personal data, account data, and survey responses into a comprehensive training dataset for the machine learning model. This combination of diverse data types provides the breadth and depth necessary for causal inference while maintaining the simplicity of automated data collection through the model.
Data Source
AI summary
A computing system is configured to generate a predictive model during training of a machine learning program using a training data set including a personal data set of a plurality of first users. The predictive model is configured to generate predicted survey data with respect to a second user by correlating a personal data set of the second user to the personal data set of at least one of the first users. The predicted survey data includes data regarding the predicted responses of the second user to a survey from which the survey data of each first user is derived, as well as one or more assessment scores calculated from the survey. The computing system is configured to take an action with respect to a user device of the second user in reaction to the generating of the predicted survey data regarding the second user.