Table 3.
The performance of screening models created using different machine learning algorithms.
| Machine learning algorithms |
MCI vs. control |
Dementia vs. control |
Cog dys vs. control |
|---|---|---|---|
| Overall accuracy (kappa) | Overall accuracy (kappa) | Overall accuracy (kappa) | |
| Screening model | 55.5% (0.051) | 55.6% (0.030) | 70.6% (0.000) |
| Logistic regression | 68.2% (0.359) | 97.2% (0.943) | 82.2% (0.551) |
| Penalized logistic regression | 72.7% (0.463) | 96.8% (0.935) | 81.9% (0.590) |
| Linear SVM | 67.8% (0.351) | 97.2% (0.943) | 80.8% (0.522) |
| Linear discriminant analysis | 70.6% (0.412) | 95.2% (0.903) | 82.7% (0.576) |
| Decision tree | 75.1% (0.499) | 97.2% (0.943) | 75.8% (0.637) |
| Radial basis function kernel SVM | 72.2% (0.436) | 97.6% (0.951) | 82.4% (0.552) |
| Random forest | 80.8% (0.618) | 97.2% (0.943) | 89.2% (0.747) |
| Gradient boosting | 93.5% (0.869) | 99.9% (1.000) | 95.5% (0.891) |
| Neural network | 76.7% (0.530) | 97.6% (0.951) | 85.6% (0.640) |
Accuracy is presented as a percentage with Cohen’s kappa in parentheses. MCI, mild cognitive impairment; Cog dys, cognitive dysfunction; SVM, support vector machine.