基于深度森林的心衰死亡率可解释预测模型

An interpretable prediction model of heart failure mortality based on deep forest

在“大数据+健康中国”背景下,针对目前ICU患者心衰死亡率预测模型中性能无法满足临床应用的要求以及缺乏可解释性的问题,提出一种基于深度森林的ICU患者心衰死亡率的可解释预测模型.在数据预处理的前提下,对其建立深度森林预测模型;与已有研究中基于bp-SVM和AB-CNN-BiLSTM等多种机器学习模型进行综合对比实验,深度森林模型的各个评价指标分别达到0.9801、0.9851、0.9802和0.9938,其结果均优于所选的对比模型,证明了该模型的有效性;利用SHAP框架增强模型的可解释性,根据可视化结果获得了相关的重要影响因素排名,为有效降低其诊断费用以及协助医护人员作出及时精确的临床诊断策略提供决策参考.

Under the background of“big data+healthy China”,an explainable prediction model of heart failure mortality in ICU patients based on deep forest was proposed to solve the problems that the performance of the current heart failure mortality prediction model for ICU patients could not meet the requirements of clinical application and the lack of interpretability.Under the premise of data preprocessing,a deep forest prediction model was established.Compared with the comprehensive comparative experiments based on various machine learning models such as bp-SVM and AB-CNN-BiLSTM in existing studies,the deep forest model reached 0.9801,0.9851,and 0.9802 and 0.9938,the results were better than the selected comparative model,which proved the effectiveness of the model,and the SHAP framework was used to enhance the interpretability of the model,and the ranking of relevant important influencing factors was obtained according to the visualization results,which provided decision-making reference for effectively reducing its diagnosis cost and assisted medical staff to make timely and accurate clinical diagnosis strategies.