Table 4.
Ablation analysis for the best performing model GP-α, τ = 1.
| Feature set | Accuracy | Precision | Recall | F1 score |
|---|---|---|---|---|
| Past observations and outcomes | 0.631 (0.015) | 0.783 (0.022) | 0.643 (0.026) | 0.706 (0.015) |
| Teaching type(s) | 0.597 (0.015) | 0.870 (0.020) | 0.599 (0.022) | 0.709 (0.015) |
| Student’s observed emotional state | 0.595 (0.023) | 0.995 (0.003) | 0.583 (0.023) | 0.735 (0.018) |
| Teacher’s communication strategy | 0.588 (0.015) | 0.784 (0.045) | 0.605 (0.026) | 0.682 (0.019) |
| Student’s attributes | 0.572 (0.025) | 0.883 (0.016) | 0.580 (0.027) | 0.700 (0.022) |
| Context | 0.572 (0.018) | 0.837 (0.021) | 0.586 (0.025) | 0.689 (0.015) |
| Teaching objective | 0.566 (0.023) | 1 (0.000) | 0.566 (0.023) | 0.722 (0.019) |
| ¬ Teacher’s communication strategy | 0.684 (0.022) | 0.791 (0.024) | 0.694 (0.028) | 0.739 (0.017) |
| ¬ Context | 0.693 (0.012) | 0.785 (0.020) | 0.707 (0.030) | 0.743 (0.010) |
| ¬ Past observations and outcomes | 0.693 (0.015) | 0.790 (0.020) | 0.704 (0.026) | 0.744 (0.014) |
| ¬ Student’s attributes | 0.695 (0.018) | 0.795 (0.016) | 0.705 (0.034) | 0.747 (0.014) |
| ¬ Student’s observed emotional state | 0.696 (0.016) | 0.784 (0.019) | 0.709 (0.030) | 0.744 (0.017) |
| ¬ Teaching type(s) | 0.696 (0.019) | 0.802 (0.022) | 0.704 (0.030) | 0.749 (0.015) |
| ¬ Teaching objective | 0.705 (0.018) | 0.794 (0.019) | 0.716 (0.030) | 0.752 (0.018) |
| All features | 0.707 (0.015) | 0.796 (0.020) | 0.718 (0.026) | 0.754 (0.014) |
A single kernel function is used (Eq. (1)) across the entire feature space to allow a straightforward interpretation of the results. The upper part of the Table enumerates accuracy estimates when a single feature category is used. At the lower part, "¬” denotes the absence of the specified feature category. For all estimates we have also included their standard deviation (in parentheses).