Table 2.
Performance of SMO, NB, and kNN under the two problem transformation methods, label powerset (LP) and binary relevance (BR).
| Dimension | Label powerset | Binary relevance | ||||
|---|---|---|---|---|---|---|
| SMO | NB | kNN | SMO | NB | kNN | |
| Behavioral domain | 0.413 29.4% | 0.374 25.0% | 0.285 14.6% | 0.437 24.1% | 0.537 23.3% | 0.350 08.5% |
| Cognitive paradigm class | 0.460 43.2% | 0.404 37.5% | 0.187 17.0% | 0.416 28.3% | 0.464 34.7% | 0.262 11.7% |
| Instruction type | 0.485 36.1% | 0.475 36.5% | 0.390 26.8% | 0.494 25.9% | 0.538 23.9% | 0.488 20.2% |
| Response modality | 0.741 54.2% | 0.733 51.0% | 0.636 48.2% | 0.740 47.4% | 0.744 49.8% | 0.698 41.7% |
| Response type | 0.704 51.4% | 0.689 51.8% | 0.619 41.6% | 0.702 44.5% | 0.715 46.5% | 0.656 33.2% |
| Stimulus modality | 0.838 78.1% | 0.842 78.1% | 0.741 68.1% | 0.816 74.9% | 0.814 72.4% | 0.768 65.2% |
| Stimulus type | 0.439 30.7% | 0.444 32.7% | 0.317 16.9% | 0.387 21.0% | 0.478 20.6% | 0.368 16.5% |
All results are based on the abstract alone corpus. Decimals are F1-micro scores and percentages are exact matches. The strict winner for each transformation-label dimension combination is highlighted. See text for details.