Layered Gradient Boosting Machine for Interpretable Insurance Pricing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models, such as gradient boosting machines, provide predictive power but lack transparency, making it difficult to interpret the effects within decision trees, especially in insurance premium rating structures where regulators require transparent algorithms for fair treatment of customers.
Innovation Solution
A layered gradient boosting machine (LGBM) model is developed that converts records to categorical variables, selects appropriate parameters and hyperparameters, and trains decision trees with different depths to compute derivatives and update model parameters, ensuring transparency and interpretability while maintaining predictive power.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If gradient boosting machines are used to improve predictive power, then predictive accuracy is improved, but transparency and interpretability deteriorate
Solution Approach 1:
The patent segments the gradient boosting model into multiple decision trees with different depths and structures. Each tree captures different aspects of the data, allowing the model to maintain high predictive accuracy while individual trees remain interpretable. The segmentation of complex computations into discrete, visualizable tree structures enables stakeholders to understand the decision-making process.
Solution Approach 2:
The patent introduces a new dimension of interpretability by visualizing decision trees in a hierarchical structure with different depths. This dimensional transformation allows the complex multi-layered computations to be represented in a way that is both computationally effective and human-readable, resolving the contradiction between accuracy and transparency.
2Productivity
If complex machine learning models are used to improve predictive power, then model efficacy is improved, but algorithm transparency deteriorates
Solution Approach 1:
The patent divides the complex machine learning model into multiple simpler decision trees, each contributing to the overall predictive power. This segmentation maintains model efficacy while reducing the apparent complexity by breaking down the algorithm into discrete, understandable components that can be individually analyzed.
Solution Approach 2:
The patent merges multiple decision trees with different depths and structures into a unified model framework. This combination allows the system to leverage the strengths of each individual tree while presenting a cohesive, interpretable structure that maintains high efficacy without excessive complexity.
3Ease of operation
If transparent algorithms are implemented to meet regulatory requirements, then customer fairness is improved, but predictive accuracy deteriorates
Solution Approach 1:
The patent segments the algorithm into visualizable decision trees that satisfy regulatory transparency requirements while maintaining predictive accuracy through the collective power of multiple trees. Each segmented tree can be individually reviewed for fairness, while the ensemble maintains high accuracy.
4Measurement precision
If decision trees with multiple computations are used to improve predictive power, then model accuracy is improved, but interpretability of effects deteriorates
Solution Approach 1:
The patent segments the complex computations into distinct decision trees with different depths, making it easier to detect and measure the effect of each computational path. Each tree represents a specific decision-making pathway that can be independently analyzed for interpretability while contributing to overall accuracy.
Data Source
AI summary
A layered machine learning system for processing data. The machine learning system comprises decision trees with different depths. An iterative training process is performed on the layered machine learning system to determine the structures of the decision trees based on prior predictions. The fitted decision trees are further configured to update leaf values with a gradient boosting method. By cumulating the predictions of decisions trees in prior iterations, interaction effects are modeled among different depths within the layered machine learning system.