Table 5.
Class-specific evaluation measures based on multiclass classification of the seven model architectures. CNN-LSTM: Convolutional neural network with a long short-term memory layer. SE: Stacking-Based Ensemble. BERT: Bidirectional Encoder Representations from Transformers
| Performance multiclass | |||||||
|---|---|---|---|---|---|---|---|
| Embedding CNN-LSTM |
Embedding SE |
Bert CNN-LSTM |
Bert SE |
Word2Vector CNN-LSTM |
Word2Vector SE |
Count vectorizer SE |
|
| Never-Smoker | |||||||
| Precision | 97% | 95% | 98% | 96% | 94% | 95% | 95% |
| Recall | 97% | 96% | 97% | 93% | 96% | 96% | 98% |
| F1-Score | 97% | 96% | 97% | 94% | 85% | 95% | 96% |
| Active-Smoker | |||||||
| Precision | 93% | 89% | 95% | 85% | 89% | 87% | 93% |
| Recall | 91% | 86% | 91% | 83% | 87% | 86% | 91% |
| F1-Score | 92% | 88% | 93% | 84% | 88% | 86% | 92% |
| Former-Smoker | |||||||
| Precision | 94% | 91% | 94% | 88% | 92% | 90% | 94% |
| Recall | 95% | 92% | 97% | 90% | 92% | 91% | 94% |
| F1-Score | 95% | 92% | 95% | 94% | 92% | 91% | 94% |