| CPC G06N 20/00 (2019.01) | 12 Claims |
|
1. An AI (artificial intelligence)-based condition classification system for patients with novel coronavirus, characterized in comprising: a classification model acquisition module, a preprocessing module, and a condition classification module;
wherein the classification model acquisition module is configured to train a plurality of binary classification models that classify a patient condition according to a patient data, and obtain one binary classification model with a highest accuracy from the binary classification models as a target model, and determine interpretable features in the patient data;
the preprocessing module is configured to extract the interpretable features in patient data to be classified, and then perform a preprocessing operation on the extracted interpretable features to fill in a missing value and replace an abnormal value among the extracted features, so as to get features to be classified after the preprocessing operation is finished;
the condition classification module is configured to use the features to be classified as inputs for the target model, and use the target model to complete a condition classification of the patient data to be classified,
wherein the classification model acquisition module comprises:
a preprocessing unit, configured to preprocess medical test data labeled with two types of data to fill in the missing value and replace the abnormal value among the data, so as to obtain a data set after preprocessing of medical test data labeled with the two types of data is completed;
a data set dividing unit, configured to divide the data set into a training set, a validation set and a test set according to a preset ratio;
each of N model training units, configured to establish each of the binary classification models that classifies the patient condition according to the patient data, set category weights, adopt the training set and the validation set to train and validate the each of the binary classification models after the category weights are set, and adopt the test set to evaluate an accuracy of each of the binary classification models after training; wherein N binary classification models established by the N model training units are different from each other, and each of the binary classification models outputs a plurality of feature importance after completing training;
a decision-making unit, configured to select one binary classification model with the highest accuracy from the N trained binary classification models as a candidate model, and select top K features with highest feature importance as the interpretable features according to the plurality of feature importance output by the candidate model;
a model retraining unit, configured to eliminate features other than the interpretable features in the training set and the validation set to obtain a new training set and a new verification set respectively, and adopt the new training set and the new validation set to retrain and revalidate the candidate model, so that after the candidate model is retrained, an optimal binary classification model is obtained and adopted as the target model; wherein N and K are both positive integers.
|