•  
  •  
 

Turkish Journal of Electrical Engineering and Computer Sciences

Author ORCID Identifier

XUE YUEWEI: 0000-0002-3410-1403

SHAOPENG GUAN: 0000-0001-9647-7396

JIA WANHAI: 0009-0004-8658-0161

DOI

10.55730/1300-0632.4108

Abstract

Effective diabetes management relies on precise prediction of blood glucose levels to minimize complications. However, the patterns and fluctuations in blood glucose vary significantly among patients, posing a challenge for existing prediction methods. Many current approaches fail to accommodate these individual differences, leading to less reliable predictions. In response to this challenge, we present LGformer, a novel prediction model based on the Informer architecture, designed to enhance both flexibility and accuracy. LGformer improves upon Informer by integrating LSTM and GRU layers into its probSparse Self-attention mechanism, allowing for personalized processing of blood glucose data tailored to each patient's unique profile. Additionally, we introduce custom weights that adjust the model’s focus on different aspects of the input data, addressing the diverse characteristics of blood glucose measurements more effectively. Experimental results on data from 16 diabetic patients demonstrate that LGformer surpasses four benchmark models, including TimesNet, in both MAE and RMSE metrics. This enhanced model successfully accommodates individual data variations, resulting in more accurate blood glucose level predictions.

Keywords

Informer, LSTM, GRU, time series forecasting

First Page

883

Last Page

905

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Download

Share

COinS