You are on page 1of 1

I.

R ESULTS
Our forecasting data analysis reveals that our model per-
forms well in simulating signal fluctuations in many instances.
However, it faces challenges when predicting large turning
points in the future. This indicates the need for further im-
provement in our forecasting model’s accuracy.
Mean Blood Pressure Heart Rate
Models Cov. MSE MAPE DTW MSE MAPE DTW
WC 18.78 7.83 20.44 7.37 6.53 13.12
NHiTS
W/o C 18.02 7.46 20.46 7.22 6.38 13.97
WC 19.79 8.03 19.37 8.73 7.38 14.36
NBeats
W/o C 18.52 7.60 17.63 7.48 6.60 10.71
WC 18.89 7.66 25.93 7.71 6.72 16.12
TFT
W/o C 19.45 7.82 25.65 8.12 6.72 16.65

TABLE I: Forecasting results.

Fig. 1. Results for all models.

Fig. 2. Results with covariates and without covariates.

In our study, we have also conducted a comparative analysis


of the performance with and without covariates. Interestingly,
we observed that the model without covariates outperformed in
the case of H-Beats , N-Hits, while the inclusion of covariates
proved more effective for TFT. These findings underscore the
importance of tailoring our modeling approach to the specific
requirements of different forecasting scenarios.
Our research demonstrates the potential of machine learning
technology in medical research, offering valuable insights for
healthcare professionals. However, it is important to acknowl-
edge the significant variability in human physical conditions.
Additionally, the presence of various potential influencing
factors, such as medication and medical procedures, neces-
sitates thorough investigation to enhance the accuracy and
applicability of our models in real-world medical settings.

You might also like