Modeling Biomedical Time Series: A Comparative Study Between PSTA-TCN and Vanilla TCN for Glucose Data Forecasting in Diabetic Patients

• Miguel A. Santos UEA
• Áurea H. da S. Melo UEA
• J. S. dos Santos UEA
• I. C. M. Gonçalves UEA
• Waldeyde O. M. Santos UEA
• Luís Cuevas R. UEA

DOI: https://doi.org/10.5753/sbcas.2025.7020

Abstract

This study compares the performance of one-dimensional convolutional layer-based models, PSTA-TCN and Vanilla TCN (keras-tcn) in one-step-ahead prediction of glucose levels in diabetic participants. Continuous glucose and heart rate measurements were used as predictor variables, and the models were evaluated using metrics such as MAE, RMSE, and training time. Continuous heart rate and glucose data were collected from 20 diabetic volunteers over a period of 3 months. The results show that Vanilla TCN achieves lower predictive error, while PSTA-TCN presents higher computational efficiency.

References

Bai, S., Kolter, J. Z., and Koltun, V. (2018). An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. arXiv preprint arXiv:1803.01271.

Fan, J., Zhang, K., Huang, Y., Zhu, Y., and Chen, B. (2023). Parallel spatio-temporal attention-based tcn for multivariate time series prediction. Neural Computing and Applications, 35(18):13109–13118.

Impedovo, D., Gattulli, V., Sarcinella, L., Veneto, D., Satalino, G., Balenzano, A., Lovergine, F., Albertini, C., Palmisano, D., Mattia, F., et al. (2024). A novel multi-speed dilated convolutional deep neural network architecture for meteorological forecasting. In 2024 IEEE International Humanitarian Technologies Conference (IHTC), pages 1–7. IEEE.

Xie, J. and Wang, Q. (2020). Benchmarking machine learning algorithms on blood glucose prediction for type i diabetes in comparison with classical time-series models. IEEE Transactions on Biomedical Engineering, 67(11):3101–3124.