Kernel-Based SVR for Censored Data Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current predictive models for time-to-event analyses, particularly those involving right-censored data, face challenges in handling high-dimensional, heterogeneous datasets effectively, as existing models like the Cox proportional hazards model are limited by regression degrees of freedom and do not adequately utilize censored data.
Innovation Solution
A kernel-based model using support vector machine regression (SVR) with a modified penalty function that differentiates between censored and non-censored data, allowing for the inclusion of right-censored data points and improving predictive accuracy by using a censored-data penalty function with a larger slack variable, and performing feature selection in a greedy-forward manner.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the Cox proportional hazards model is used to analyze censored data, then the model can properly process right-censored data, but the model is limited by regression degrees of freedom and cannot effectively handle high dimensional datasets
Solution Approach 1:
The patent transforms the traditional Cox model parameters by introducing a kernel function that maps input features into a higher-dimensional space, changing the parameter representation from direct feature coefficients to kernel-based implicit features. This allows the model to handle high-dimensional data without increasing the number of explicit parameters, resolving the contradiction between processing capability and model complexity.
Solution Approach 2:
The patent applies dimensionality change by using kernel functions to implicitly map data into a higher-dimensional feature space. The kernel trick allows computations in this higher-dimensional space to be performed efficiently using only the original input data, enabling the model to capture complex high-dimensional relationships without the computational burden of explicitly representing all dimensions.
2Device complexity
If Support Vector Machines are used to handle high dimensional datasets, then the model performs well with high dimensional data, but the model is not well-suited for use with censored data
Solution Approach 1:
The patent merges the strengths of Support Vector Machines (handling high-dimensional data through kernel methods) with the strengths of the Cox proportional hazards model (proper handling of censored data through likelihood-based estimation). The resulting kernel-based Cox model combines the kernel trick for dimensionality handling with the partial likelihood framework for censored data, achieving both objectives simultaneously.
3Device complexity
If traditional regression models are used, then the model structure is simple, but the model cannot effectively utilize both censored and uncensored data points in high dimensional datasets
Solution Approach 1:
The patent segments the data into censored and uncensored portions and applies different penalty functions to each segment in the loss function. This segmentation allows the model to handle each type of data point appropriately while maintaining a unified framework, improving predictive accuracy without requiring completely separate models for each data type.
Data Source
AI summary
A method of producing a model for use in predicting time to an event includes obtaining multi-dimensional, non-linear vectors of information indicative of status of multiple test subjects, at least one of the vectors being right-censored, lacking an indication of a time of occurrence of the event with respect to the corresponding test subject, and performing regression using the vectors of information to produce a kernel-based model to provide an output value related to a prediction of time to the event based upon at least some of the information contained in the vectors of information, where for each vector comprising right-censored data, a censored-data penalty function is used to affect the regression, the censored-data penalty function being different than a non-censored-data penalty function used for each vector comprising non-censored data.