TABLE 4.
Mean accuracy of classifying pictures of low and high ratings for valence and arousal in each dataset (SVM-RBF classifier with 10-fold cross-validation).
| Dataset | Mean | 95% CI | t1 | p1 |
| Valence | ||||
| IAPS | 58.2% | 52.5–63.9% | 3.44 | 0.0101 |
| GAPED | 65.1% | 55.2–75.0% | 3.62 | 0.0074 |
| NAPS-H | 64.3% | 58.1–70.5% | 5.23 | 0.0005 |
| NAPS-SFIP | 59.5% | 51.2–67.8% | 2.59 | 0.0291 |
| OASIS | 58.7% | 55.3–62.0% | 5.83 | 0.0003 |
| DIRTI | 75.5% | 66.2–84.8% | 6.20 | 0.0002 |
| Arousal | ||||
| IAPS | 59.3% | 50.0–59.3% | 3.82 | 0.0041 |
| GAPED | 62.4% | 52.7–72.1% | 2.90 | 0.0175 |
| NAPS-H | 57.4% | 52.6–62.3% | 3.45 | 0.0073 |
| NAPS-SFIP | 62.0% | 59.0–65.1% | 8.85 | <0.0001 |
| OASIS | 57.5% | 51.8–63.2% | 2.97 | 0.0157 |
| DIRTI | 71.5% | 63.1–79.9% | 5.76 | 0.0003 |
The by-chance classification accuracy is 50% for this binary task. 1 t statistics and p values for difference to random accuracy rate of 50% (two-sided one sample t-test, df = 9).