Classifier Accuracy via Dual Threshold Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Classifier models used in medical applications face challenges in achieving high accuracy, especially when further training data is not readily available, limiting the improvement of classifier model performance.
Innovation Solution
The method involves obtaining two datasets, training a classifier using one dataset, and then using the trained classifier to provide two classifications for input data from the second dataset, with different predictive thresholds applied to optimize accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If further training data is collected to improve classifier model performance, then accuracy improves, but data collection complexity and time increase
Solution Approach 1:
The patent applies preliminary action by using a validation dataset to pre-determine optimal predictive thresholds before the classifier is deployed on new data. The validation dataset is used to train a threshold determination model that predicts the best thresholds for achieving high accuracy, avoiding the need to collect and process additional training data after the main classifier is trained. This preliminary threshold determination based on validation data resolves the contradiction by eliminating post-training data collection requirements.
Solution Approach 2:
The patent uses a copy of the validation dataset to train a threshold determination model. Instead of using the original validation data directly for final threshold selection, a copy is used to train a predictive model that can determine thresholds for new datasets. This copying approach allows the system to prepare threshold determination capabilities in advance without requiring access to the original validation data during deployment, thereby avoiding additional data collection needs.
2Measurement precision
If a single predictive threshold is used for classification, then the classification process is simple, but both sensitivity and specificity cannot be optimized simultaneously
Solution Approach 1:
The patent segments the threshold determination process into two distinct stages: first, training a threshold determination model using validation data; second, applying this model to determine optimal thresholds for new datasets. This segmentation allows the complex task of optimizing multiple thresholds to be broken down into a manageable training phase followed by automated application, resolving the contradiction between achieving high accuracy and maintaining process simplicity.
Solution Approach 2:
The patent implements feedback by using the validation dataset to train a threshold determination model that learns from the relationship between input data and optimal threshold values. This feedback mechanism allows the system to automatically adjust thresholds based on patterns learned from validation data, achieving optimized sensitivity and specificity without requiring complex manual threshold determination processes.
Data Source
AI summary
A computer implemented method, comprising: obtaining a first dataset comprising first input data corresponding to a first class; obtaining a second dataset comprising second input data corresponding to the first class; training at least one classifier using the first dataset; inputting the second input data from the second dataset to the at least one classifier, and providing a classification model comprising a first classification and a second classification, wherein the first classification predicts a greater proportion of the second input data corresponding to the first class to be in the first class than the second classification.