Contextual Discrete Choice Modeling With Fixed Subject Factors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing discrete choice experiments face challenges in predicting subject preferences without sufficient data for model training, particularly in capturing the contributions of profile and subject factors effectively.
Innovation Solution
A computer-program product generates a fixed-factor model using a predictive model with weighted terms, where a first subject factor is fixed, allowing for predictions in environments without model training data, and adjusts weights for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a predictive model is trained using discrete choice experiment data, then the model can capture contributions of profile and subject factors, but the model requires sufficient training data which is not always available
Solution Approach 1:
The system pre-computes and stores summary statistics and model parameters from previous experiments or databases. When a new prediction is needed, the system retrieves pre-computed information rather than performing full model training from scratch, enabling accurate predictions without requiring sufficient new training data.
Solution Approach 2:
The system creates a simplified representation or copy of the full predictive model that captures essential patterns from training data. This compressed model structure allows for new predictions to be made using a subset of the original training data, maintaining predictive accuracy while reducing data requirements.
2Adaptability or versatility
If the predictive model includes multiple factors (profile and subject factors), then the model can capture more complex preferences, but the model complexity increases
Solution Approach 1:
The system divides the complex predictive model into separate modular components: profile factor parameters, subject factor parameters, and interaction terms. Each component can be independently estimated, stored, and retrieved, allowing the system to handle complex multi-factor models while maintaining computational efficiency and manageability.
Solution Approach 2:
The system pre-estimates and stores parameter values for multiple factors from previous data collection. When making predictions, the system retrieves these pre-computed parameters rather than performing complex calculations in real-time, enabling the model to handle multiple factors without increasing operational complexity.
3Measurement precision
If the system generates predictions for specific subject groups, then the predictions can be more targeted and accurate, but the system requires additional data processing and model adjustments
Solution Approach 1:
The system pre-categorizes subjects into groups based on demographic characteristics and pre-computes specific model parameters for each group. When a prediction is needed for a specific subject group, the system retrieves the pre-computed group-specific parameters, enabling targeted predictions without requiring additional data processing or model retraining.
Solution Approach 2:
The system maintains different model parameter sets for different subject groups, allowing each group to have customized predictions based on their specific characteristics. This localized approach enables accurate group-specific predictions while avoiding the need to reprocess the entire dataset for each prediction request.
Data Source
AI summary
A computing system obtains multiple factors for a first environment in a first experiment. The multiple factors include one or more profile factors and one or more subject factors. Each profile factor of the one or more profile factors specifies multiple candidate features for the first environment. Each subject factor of the one or more subject factors specifies multiple candidate levels categorizing different subjects interacting with the first environment. The computing system obtains a predictive model for predicting probabilities of multiple candidate outcomes for the first environment. The predictive model has weighted model terms. The computing system receives a selection of a first level for a first subject factor of the one or more subject factors. The computing system generates a computer-generated representation of a fixed-factor model. The fixed-factor model predicts multiple candidate outcomes for a second environment with subjects defined by the selection of the first level.