M.S. AAI Capstone Chronicles 2024

16

models. Thus, it was decided to focus solely on comparing the pre-trained DistilBERT model with the custom transformer model. Validation Dataset The evaluation results on the validation dataset for the pre-trained DistilBERT and custom transformer models can be seen in Figure 8. Figure 8

Note: Results Comparison on Validation Dataset As mentioned previously, the custom transformer was trained on the entire dataset, and the pre-trained DistilBERT model only on a subset due to resource constraints. When comparing between the two, the DistilBERT model shows better accuracy, recall, and F1 score, whereas the custom transformer has better precision.

66