基于改进残差网络和SHAP的糖尿病预测及可解释性分析

魏国政1; 魏丽丽2; 宋廷强1; 渠蓉蓉1; 孙媛媛3; 董凡琦1

基于改进残差网络和SHAP的糖尿病预测及可解释性分析

引用本文：魏国政1，魏丽丽2，宋廷强1，渠蓉蓉1，孙媛媛3，董凡琦1.基于改进残差网络和SHAP的糖尿病预测及可解释性分析[J].计算技术与自动化,2026,(1):151-157

摘要点击次数: 243

全文下载次数: 220

作者	单位
魏国政1，魏丽丽2，宋廷强1，渠蓉蓉1，孙媛媛3，董凡琦1	(1.青岛科技大学信息科学技术学院,山东青岛 266061；2.青岛大学附属医院院长办公室,山东青岛 266000；3.青岛科技大学数据科学学院,山东青岛 266061)

中文摘要:针对糖尿病预测领域中可靠性与可解释性不足问题，提出了基于改进深度残差网络的预测算法。该算法嵌入了根据数据集特性设计的特征自注意力机制，并辅以SHAP模型以增强可解释性。SHAP能够精准定位并可视化影响糖尿病预测的关键因素，提升预测逻辑的透明度与实用价值。实验在Pima公开数据集及青岛某三甲综合医院私有数据集上展开，RAC模型与朴素贝叶斯、逻辑回归、支持向量机等模型进行了对比。结果显示，RAC的分类准确率、灵敏度、特异性、F1分数值均优于其他模型，验证了其在临床实践中早期预警或辅助诊断的潜力。

中文关键词:糖尿病预测可解释性改进深度残差网络特征自注意力机制 SHAP模型

Diabetes Prediction and Interpretability Analysis Based on Improved Residual Network and SHAP

Abstract:To address the lack of reliability and interpretability in the field of diabetes prediction, a prediction algorithm based on improved deep residual network is proposed. The algorithm embeds a feature self-attention mechanism designed according to the characteristics of the dataset, and is complemented by the SHAP model to enhance the interpretability, which can pinpoint and visualise the key factors affecting the prediction of diabetes mellitus, and enhance the transparency and practical value of the prediction logic. The experiments were carried out on the public dataset of Pima and the private dataset of a tertiary general hospital in Qingdao, and the RAC model was compared with the plain Bayes, logistic regression, and support vector machine models. The results show that the classification accuracy, sensitivity, specificity, and F1 score values of RAC are better than those of other models, validating its potential for early warning or assisted diagnosis in clinical practice.

keywords:diabetes prediction interpretability improved deep residual network feature self-attention mechanism SHAP model

查看全文 查看/发表评论 下载pdf阅读器