Table 2.
Performance of the proposed methodology and global approaches on the complex-valued fMRI dataset. This table presents the classification accuracy (first row) and the sensitivity/specificity rates (second row) of our local-oriented methodology using ν-MKL lp-norm MKL and SVM for single-source data (magnitude or phase) and different source combination approaches. It also shows the results obtained by global approaches. Notice that SVM is applied to both the proposed approach and global approaches. The reported values are attained by these algorithms using linear kernels, except where noted.
| Classifier | Single Sources | Combined Sources | |||||||
|---|---|---|---|---|---|---|---|---|---|
|
| |||||||||
| Prop. Approach | Global Approach | Proposed Approach | Global Approaches | ||||||
|
| |||||||||
| Magn. | Phase | Magn. | Phase | Comb. 1 | Comb. 2 | Comb. 3 | Whole Data | Filt. Data | |
| SVM | 0.77 | 0.64 | 0.62 | 0.58 | 0.80 | 0.79 | 0.79 | 0.63 | 0.80 |
| 0.84/0.67 | 0.65/0.64 | 0.71/0.48 | 0.55/0.62 | 0.85/0.71 | 0.82/0.74 | 0.82/0.74 | 0.71/0.50 | 0.82/0.76 | |
|
| |||||||||
| Global | – | – | 0.76 | 0.61 | – | – | – | 0.80 | – |
| RFE-SVM | – | – | 0.81/0.69 | 0.63/0.57 | – | – | – | 0.92/0.62 | – |
|
| |||||||||
| ν-MKL | 0.80 | 0.70 | – | – | 0.76 | 0.76 | 0.85 | – | – |
| (linear) | 0.85/0.71 | 0.69/0.71 | – | – | 0.82/0.67 | 0.84/0.64 | 0.90/0.76 | – | – |
|
| |||||||||
| ν-MKL | 0.78 | 0.68 | – | – | 0.68 | 0.77 | 0.85 | – | – |
| (Gaussian) | 0.84/0.69 | 0.71/0.64 | – | – | 0.77/0.55 | 0.87/0.62 | 0.92/0.74 | – | – |
|
| |||||||||
| lp-norm | 0.78 | 0.64 | – | – | 0.76 | 0.72 | 0.84 | – | – |
| MKL | 0.84/0.69 | 0.66/0.62 | – | – | 0.82/0.67 | 0.73/0.71 | 0.90/0.74 | – | – |