Predictive Model Framework Using Region-Specific Feature Engineering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Supervised learning models often exhibit varying accuracy across different target value intervals, and feature engineering can introduce noise if additional features have low correlation with target values, leading to suboptimal performance in certain regions.
Innovation Solution
The approach involves assigning record-specific engineered features based on model regions determined for each input record, using SHAP values to identify feature contributions, and enhancing training data with region-specific features to improve predictive model accuracy while minimizing noise and data complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional feature engineering is applied to improve overall model accuracy, then the Mean Absolute Error decreases, but the model exhibits inconsistent accuracy across different target value intervals with some regions experiencing noise suppression
Solution Approach 1:
The patent segments the feature engineering process by creating region-specific engineered features based on target value intervals. Instead of applying uniform feature engineering across all data, the system divides the target space into regions and generates tailored features for each region, allowing the model to achieve consistent accuracy across different intervals while maintaining overall predictive accuracy
Solution Approach 2:
The patent implements local quality by making feature engineering adaptive to local characteristics of different target value regions. The system generates different engineered features for different regions based on local data patterns and correlations, ensuring that each region receives features optimized for its specific characteristics rather than applying a one-size-fits-all approach
2Measurement precision
If additional features are added to improve model accuracy, then the predictive performance improves in some regions, but noise is introduced when the additional features have low correlation with target values
Solution Approach 1:
The patent changes the parameters of feature engineering by dynamically adjusting which features are created and how they are constructed based on the target value region. The system evaluates feature correlation with target values within each region and selectively generates features only when they demonstrate sufficient correlation, thereby improving accuracy in relevant regions while avoiding noise introduction from low-correlation features
Data Source
AI summary
A system and method include input of data records to a first trained predictive model to obtain a predicted value associated with each input data record. A model region is then associated with each of the input data records based on the first trained predictive model, the input data records and the predicted values. Enhanced input data records are generated by, for each model region, adding derived values of engineered features associated with the model region to input data records associated with the model region and default values of the engineered features associated with the model region to input training records not associated with the model region. The enhanced input data records are input to a second trained predictive model to obtain an enhanced predicted value associated with each input data record.