Classifier Decision Support Using Clustered Convolutional Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional classification technologies struggle with high misclassification error rates in scenarios with small training datasets, admixture of genotype or phenotype subgroups, high dimensionality, and unbalanced feature values, leading to inaccurate predictions and model calibration issues.
Innovation Solution
Implement a clustering process to identify statistical dependencies among variables, apply convolution transformations to reduce dimensionality, and use deep learning neural networks to generate a classification model, enhancing signal-to-noise ratio and improving pattern recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional classification methods are used, then the classification process is simple, but misclassification error rates are high in scenarios with small training datasets, high dimensionality, and unbalanced feature values
Solution Approach 1:
The patent segments the high-dimensional feature space by identifying and grouping variables into statistically significant clusters. This segmentation reduces the complexity of the classification task by organizing features into meaningful groups rather than treating all features individually, thereby improving classification accuracy without proportionally increasing complexity
Solution Approach 2:
The patent applies convolution transformations to the clustered variable matrix, effectively reducing dimensionality by operating on local patterns within clusters rather than the full high-dimensional space. This dimensional transformation maintains important relationships while reducing the computational burden and improving reliability
2Device complexity
If the dimensionality of the input feature space is reduced, then the classification complexity decreases, but information loss may occur
Solution Approach 1:
By segmenting features into statistically significant clusters, the patent preserves local relationships and patterns within each cluster while reducing the global dimensionality. This ensures that information is not lost but reorganized in a more manageable structure
Solution Approach 2:
The patent merges variables into clusters based on statistical relationships, combining redundant or correlated features while preserving the essential information through the cluster structure. This merging reduces dimensionality without proportional information loss
3Reliability
If clustering and convolution transformations are applied, then misclassification errors are reduced, but the processing time and computational resources increase
Solution Approach 1:
The patent performs clustering and convolution transformations as preliminary steps before the actual classification. By pre-processing the data into clustered matrices and applying convolution filters beforehand, the system reduces the computational burden during classification, thereby reducing overall processing time while maintaining high accuracy
Data Source
AI summary
Technologies are provided for an improved classifier apparatus and processes for improving the accuracy of classification technology including example applications of such classifiers. A process includes applying clustering to variables contributing to the classification task. The clusters may be represented in a 1-dimensional, 2-dimensional, or 3-dimensional matrix that is a spatial abstraction of the interrelationships. A convolutional transformation may be applied to the matrix so as to reduce the effective dimensionality of the classification problem and improve the signal-to-noise ration. A deep learning neural network method may be applied to the transformed network to generate an improved classification model, which may be utilized by a decision support tool. One embodiment comprises a decision support tool for detecting risk of venous thrombosis and venous thromboembolism (VTE) in a patient, based on phenotype and genomics information.