Table 4.
Comparison of different missing value imputation methods and their effect on weighted classification accuracy for Parkinson’s disease and schizophrenia. In all cases, 100 CpG sites with the highest importance values were dropped.
| Model | Parkinson’s disease | Schizophrenia | |
|---|---|---|---|
| GSE72774 | GSE152027 | GSE116379 | |
| Original (n o missed data) | 0.96 | 0.75 | 0.7 |
| Mean | 0.78 | 0.65 | 0.67 |
| Median | 0.81 | 0.59 | 0.55 |
| Mode | 0.93 | 0.55 | 0.54 |
| Random | 0.86 | 0.62 | 0.54 |
| Chained equation | 0.89 | 0.68 | 0.61 |
| Expectation maximization | 0.87 | 0.67 | 0.54 |
| KNN (K = 1) | 0.96 | 0.65 | 0.67 |
| KNN (K = 2) | 0.9 | 0.65 | 0.67 |
| KNN (K = 3) | 0.86 | 0.65 | 0.67 |