Effective physiological feature selection and medical causal reasoning method based on interpretable machine learning

Abstract

The invention discloses an effective physiological feature selection and medical causal reasoning method based on interpretable machine learning. The method comprises the following steps: acquiring medical data from an electronic medical record; decoupling the feature space into a combination of a plurality of effective features through a plurality of feature selection methods; different feature selection methods are compared, and the rationality of adopting the SHAP value in the causal reasoning field is explained; the model features are evaluated based on the SHAP, and the association between the feature space and the prediction result is analyzed; causal information is incorporated into a feature space, and an interpretable machine learning model is constructed; according to different causal models, various Shapley Values are used for providing reasonable explanations; and outputting the importance degree of each feature, the contribution degree to the sample and the causal relationship with the prediction result. According to the method, disease development is explained and disease reasoning is carried out according to effective characteristics, and the effect and interpretability of the model and the accuracy of disease diagnosis are improved.

Description

technical field [0001] The invention relates to the fields of computer and medicine, in particular to an effective physiological feature selection and medical causal reasoning method based on interpretable machine learning. Background technique [0002] Today, with the rapid development of artificial intelligence, the field of intelligent medical care has also developed rapidly. Among them, intelligent disease reasoning and diagnosis technology based on medical electronic medical records has been widely used in the medical field. [0003] Medical big data has the characteristics of large data volume, high dimensionality, high noise, diversity, complexity, incomprehensibility, uncertainty, and unbalanced data distribution. Among them, electronic medical records are a very important part of medical big data. It includes multiple pathological characteristics of patients, disease history data, examination results and other types of text information, and records the effect track...

AI Technical Summary

Problems solved by technology

The process of learning feature expression is usually a "black box model", which cannot show the logical rules and associations of its operation, that is, it lacks interpretability

For structured data and classification tasks, integrated models often have better results. For integrated learning methods, although the effect is good, it has not been able to solve the problem of interpretability.

So for a specific sample, we have no way of knowing whether each eigenvalue in this sample is really effective, and how it affects the final result

Images