Table 3.
Cross validation performance comparison of linear regression, ridge regression, PLS, ℓ2,1-norm, and G-SMuRFS: performance is measured by the root mean squared error (RMSE) and correlation coefficient (CC) between the actual and predicted scores of the test subjects. The average (avg) and standard deviation (std) of performance measures across five cross-validation trials are shown as “avg±std” for each experiment.
| (a) Performance comparison using RMSE | ADAS | MMSE | RAVLT | ||
|---|---|---|---|---|---|
| TOTAL | T30 | RECOG | |||
| G-SMuRFS | 0.7663±0.0375 | 0.8325±0.0399 | 0.8490±0.0654 | 0.8776±0.0701 | 0.9167±0.0471 |
| ℓ2,1-norm | 0.7631±0.0346 | 0.8322±0.0411 | 0.8464±0.0697 | 0.8807±0.0657 | 0.9215±0.0440 |
| PLS | 0.8844±0.0425 | 0.8999±0.0453 | 0.9246±0.0533 | 0.9515±0.0578 | 0.9706±0.0512 |
| Ridge | 0.7859±0.0327 | 0.8579±0.0332 | 0.8453±0.0697 | 0.8977±0.0553 | 0.9315±0.0430 |
| Linear | 1.0524±0.0559 | 1.1342±0.0739 | 1.1726±0.0956 | 1.3042±0.0803 | 1.2775±0.1168 |
| (b) Performance comparison using CC | ADAS | MMSE | RAVLT | ||
|---|---|---|---|---|---|
| TOTAL | T30 | RECOG | |||
| G-SMuRFS | 0.6438±0.0258 | 0.5552±0.0078 | 0.5277±0.0539 | 0.4753±0.0591 | 0.3985±0.0533 |
| ℓ2,1-norm | 0.6468±0.0175 | 0.5548±0.0058 | 0.5301±0.0544 | 0.4692±0.0522 | 0.3889±0.0588 |
| Partial Least Square | 0.4608±0.0551 | 0.4339±0.0406 | 0.3782±0.0591 | 0.3037±0.0565 | 0.2390±0.0382 |
| Ridge | 0.6167±0.0171 | 0.5127±0.0272 | 0.5296±0.0540 | 0.4364±0.0416 | 0.3632±0.0681 |
| Linear | 0.4902±0.0544 | 0.3779±0.0246 | 0.3685±0.1005 | 0.2958±0.1359 | 0.2533±0.0726 |