Early sepsis prediction based on time series and KA-Transformer models
ZHU Yu
ZHANG Tianyi
ZHANG Li
CHENG Yunzhang
Abstract：To achieve early prediction of sepsis,we designed a prediction model KA-Transformer based on time series data.A ker-nel attention mechanism was introduced in the KA-Transformer to improve issues such as limited training samples,numerous parame-ters,and uneven sample distribution.With the input of continuous time series data,at three prediction time points 1,6,and 12 h before sepsis onset,the area under the receiver operating characteristic curve of the model for predicting sepsis were 0.962,0.944 and 0.984,respectively,the accuracy rates were 92.3％,93.9％and 96.1％,respectively.The experimental results show that the KA-Transformer significantly outperforms existing methods in terms of accuracy and generalization ability for sepsis prediction,and has the potential to enhance the timeliness and reliability of prediction for sepsis prediction.
Keywords：SepsisTime seriesDeep learningKernel attentionTransformer
Publication Date：2025-04-30
Online Publishing Date：2026-09-11(First online date of this platform, not the publication date of the document)
Pages：7( 90-96 )