Table 1 The LOOCV testing correlation (\(\rho\)) and MAE of the proposed deep models and Gradient Tree Boosting are reported for single models and the ensemble of two or three models of the deep models
From: Ensemble deep model for continuous estimation of Unified Parkinson’s Disease Rating Scale III
Method | \(\rho\) | MAE | |
|---|---|---|---|
Single | Gradient Tree Boosting | 0.61 | 7.85 |
Dual-channel LSTM, hand-crafted features | 0.62 | 7.50 | |
Dual-channel LSTM, hand-crafted features, with transfer learning | 0.67 | 6.85 | |
1D CNN-LSTM for raw signals | 0.70 | 6.93 | |
2D CNN-LSTM for time–frequency data | 0.67 | 7.11 | |
Ensemble | Dual-channel LSTM, hand-crafted features, with transfer learning 1D CNN-LSTM for raw signals | 0.77 | 6.04 |
Dual-channel LSTM, hand-crafted features, with transfer learning 2D CNN-LSTM for time–frequency data | 0.76 | 5.99 | |
1D CNN-LSTM for raw signals 2D CNN-LSTM for time–frequency data | 0.74 | 6.54 | |
Dual-channel LSTM, hand-crafted features, with transfer learning 1D CNN-LSTM for raw signals 2D CNN-LSTM for time–frequency data | 0.79 | 5.95 | |