Fig. 7 | Scientific Reports

Fig. 7

From: Adaptive focal loss with personality stratification for stably mitigating hard class imbalance in multi-dimensional personality recognition

Fig. 7

The trend of the FBCE loss hyperparameter results in average accuracy (both BA and RA) and epochs (AE) across dimensions in the Essays and Kaggle datasets. The dotted lines for 7(a) and 7(b) show the baseline performance (BA, RA, and AE) of regular BCE and WBCE-S for 7(c) and 7(d). In 7(a), the FBCE-F focal factor shows an improvement in accuracy (RA and BA) regardless of the values chosen in a balanced dataset. In 7(b), the FBCE-F focal factor shows sensitivity to the hyperparameter chosen in an imbalanced dataset. In 7(c), the FBCE-W focal factor demonstrates an improvement in accuracy (RA and BA), regardless of the values chosen for the weighted factor. In 7(d), the FBCE-W focal factor shows that the model is sensitive to the \(\gamma\) in an imbalanced dataset; a well-chosen hyperparameter is vital for optimum performance.

Back to article page