Skip to main content
. 2024 Dec 28;14:31269. doi: 10.1038/s41598-024-82681-4

Table 8.

The performance after removing the deep learning module.

Model BERT-CRF ALBERT-CRF RoBERTa-CRF
Precision Recall F1 (%) Precision Recall F1 (%) Precision Recall F1 (%)
Cheating 82.56% 93.42% 87.65 37.62% 50.00% 42.94 96.10% 97.37% 96.73
Mixed 98.50% 99.85% 99.17 95.38% 97.26% 96.31 100.00% 100.00% 100
Out-listed 92.90% 97.52% 95.15 76.71% 86.54% 81.33 97.05% 98.95% 97.99
Not fresh 80.28% 90.48% 85.07 64.00% 76.19% 69.57 89.23% 92.06% 90.62
Frozen 84.80% 80.30% 82.49 64.75% 68.18% 66.42 82.11% 76.52% 79.22
Spoilage 88.19% 95.73% 91.80 66.90% 82.91% 74.05 91.13% 96.58% 93.78
Over-loading 97.67% 98.88% 98.27 87.03% 91.13% 89.04 99.89% 100.00% 99.94
Deep-processing 75.48% 79.60% 77.48 55.56% 66.09% 60.37 83.38% 85.06% 84.21
F 89.64 72.50 92.81