Table 6 Regression results for predicting the DaT ratios corresponding to the four regions of interest using the XGBoost models

From: Smartphone-based prediction of dopaminergic deficit in prodromal and manifest Parkinson’s disease

Model

Right Putamen

Left Putamen

Right Caudate

Left Caudate

Naïve Benchmark

RMSE: 0.74 ± 0.42

RMSE: 0.79 ± 0.47

RMSE: 0.60 ± 0.37

RMSE: 0.65 ± 0.43

MDS-UPDRS-III

RMSE: 0.57 ± 0.38

RMSE: 0.66 ± 0.39

RMSE: 0.55 ± 0.33

RMSE: 0.61 ± 0.39

Smartphone features

RMSE: 0.63 ± 0.33

RMSE: 0.67 ± 0.38

RMSE: 0.63 ± 0.37

RMSE: 0.59 ± 0.36

Smartphone features + MDS-UPDRS-III

RMSE: 0.49 ± 0.30

RMSE: 0.57 ± 0.34

RMSE: 0.52 ± 0.31

RMSE: 0.55 ± 0.37

Combination prediction (mean)

RMSE: 0.64 ± 0.37

RMSE: 0.69 ± 0.39

RMSE: 0.57 ± 0.34

RMSE: 0.60 ± 0.35

  1. RMSE: root mean squared error, presented as mean ± standard deviation.
  2. The naïve benchmark issues the mean of the training data as a prediction for the entire test data. This benchmark serves as a reference point against which more sophisticated machine learning methods can be compared. Note: lower RMSE values are better. The best performing model is highlighted in bold.
Back to article page