In this notebook I conduct an exploratory data analysis on the text data. Then, I train three models for fake news detection. Data is taken from this kaggle competition.
The models are:
- The pretrained DistilBERT and ALBERT Base v1 models using implementations from the
Hugging Faceframework. I used transfer learning to retrain these two models for fake news detection. - XGBoost Classifier from the
XGBoostlibrary withTF-IDFfeatures.
| Model | Dataset | Accuracy | F1-Score |
|---|---|---|---|
| DistilBERT | Validation | 0.9996 | 0.9996 |
| Test | 0.9998 | 0.9998 | |
| ALBERT Base v1 | Validation | 0.999 | 0.999 |
| Test | 0.9998 | 0.9998 | |
| XGBClassifier | Validation | 0.975 | 0.975 |
| Test | 0.973 | 0.973 |