Interpretable Machine Learning Model Mimicking Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning models, particularly deep neural networks, are difficult to interpret in healthcare predictions due to their complexity, making it challenging for clinicians to understand the reasoning behind predictions, which leads to reliance on less accurate but interpretable models like decision trees that may overfit and perform poorly on heterogeneous electronic health records.
Innovation Solution
A method is developed to create an interpretable machine learning model by training a neural network to predict health information and then using the learned features to train a gradient boosting tree or similar model to mimic the neural network's predictions, providing both accuracy and interpretability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep neural networks are used for healthcare predictions, then prediction accuracy is improved, but model interpretability deteriorates
Solution Approach 1:
The patent creates a mimic model that copies the prediction behavior of the deep neural network. The mimic model is trained to reproduce the DNN's predictions on the same input data, thereby transferring the accurate prediction capability to an interpretable model structure that clinicians can understand and trust.
Solution Approach 2:
The patent segments the modeling process into two distinct phases: first training a deep neural network for accurate predictions, then training a separate mimic model to replicate those predictions. This segmentation allows each model to serve its specific purpose - the DNN for accuracy and the mimic model for interpretability.
2Device complexity
If interpretable models like decision trees are used, then model interpretability is improved, but prediction accuracy deteriorates
Solution Approach 1:
The mimic model copies the predictive capabilities of the deep neural network while using an interpretable structure. By training the mimic model to reproduce DNN predictions, it achieves accuracy comparable to complex models while maintaining the simplicity and interpretability of tree-based structures.
3Measurement precision
If deep neural networks are used on heterogeneous EHR data, then prediction accuracy is improved, but model reliability deteriorates due to overfitting
Solution Approach 1:
The mimic model copies the generalization capabilities of the deep neural network. By training on the same heterogeneous EHR data and targeting the DNN's predictions, the mimic model learns robust patterns without the overfitting issues that plague complex models on noisy medical data.
Data Source
AI summary
A method for creating an interpretable model for healthcare predictions includes training, by a deep learning processor, a neural network to predict health information by providing training data, including multiple combinations of measured or observed health metrics and corresponding medical results, to the neural network. The method also includes determining, by the deep learning processor and using the neural network, prediction data including predicted results for the measured or observed health metrics for each of the multiple combinations of the measured or observed health metrics based on the training data. The method also includes training, by the deep learning processor or a learning processor, an interpretable machine learning model to make similar predictions as the neural network by providing mimic data, including combinations of the measured or observed health metrics and corresponding predicted results of the prediction data, to the interpretable machine learning model.