Integration of a Heart Rate Prediction Model into a Personal Health Record to Support the Telerehabilitation Training of Cardiopulmonary Patients - Biomedical Engineering Systems and Technologies - page 307

submodels were then concatenated to a complete model for one training session (see
Fig. 2). This also simplified the comparison to the real HR of the training sessions
used for validation.

2.3

Model Evaluation

To determine the quality of our model and to prevent overfitting, we performed a 2-
fold cross-validation. We divided the dataset into two parts d 0 and d 1 . Both parts were
of the same size and contained randomly selected training sessions ( n =334) from the
dataset. First, we used d 0 to train the model and validated it against the d 1 dataset then
we performed this procedure vice versa.
We calculated the root mean square error (RMSE) which quantifies the deviation
between measured and predicted heart rate over a whole training.
It is not easy to determine, which predictor of the resulting model explains which
part of the response variable, as each added predictor depends on the former one. To
uncover which predictors are of and have to be stored in the PHR, we measured the
percental improvement of the RMSE when a predictor is added to the model in rela-
tion to the former one.

Table 1. Mean contribution of the predictors on the scenario (S1-S5) model. All values
represent the improvement of the former RMSE in percent by addition of a predictor during
stepwise regression. The “-” symbol denotes that a predictor is not available in the given
scenario. The calculated average influence of a predictor is shown in column “Overall”. The
order of these values is additionally illustrated by a rank order in the last column.

Predictor S1 S2 S3 S4 S5 Overall Rank

Age 11.032 11.032 11.112 9.002 8.018 10.040 3

Gender 0.754 0.754 0.745 0.156 0 0.482 8

Load 0.368 0.368 5.556 0.646 0.078 1.403 6

Overall training duration 0.065 0.065 0 0 0 0.026 12

Duration of current training phase 0.040 0.040 0.015 0.019 0.011 0.025 13

Air pressure - 0.023 0.013 0.141 0.136 0.079 11

Temperature - 0 0 0 0 0 -

Humidity - 0 0 0.059 0 0.015 14

Resting HR - - 40.990 7.154 5.268 17.804 2

Resting BP systolic - - 0.494 0.118 0.119 0.244 9

Resting BP diastolic - - 0 0 0.086 0.029 11

Average HR of former phase - - - 57.064 54.3 55.682 1

Load phase BP systolic - - - 0 0.005 0.003 16

HR at the end of former phase - - - 0 0.027 0.013 15

Load phase BP diastolic - - - 0 0 0 -

Average HR of current phase - - - - 5.648 5.648 4

Average HR of load phase - - - - 3.548 3.548 5

Recovery pulse - - - - 0.683 0.683 7

Recovery BP diastolic - - - - 0.122 0.122 10

Borg value - - - - 0 0 -

Average HR of all phases - - - - 0 0 -

Average of all BP values systolic - - - - 0 0 -

Average of all BP values diastolic - - - - 0 0 -

Recovery BP systolic - - - - 0 0 -

Total number of predictors 5 8 11 15 24