Comparison of Accuracy of Gradient Boosted Trees in Fake News Detection

Due to the rapid expansion of the internet, more fake news are appearing every minute which is a great issue that can cause damage to society. There were many papers before that focused on the possibility of detection of fake news using different algorithms. The gradient boosted models stood out in the accuracies because of how they were able to perform well on different datasets. This paper instead highlights the differences between gradient boosted models involving XGBoost, CatBoost and lightGBM. Each of the models was tested with different hyperparameters that were fitted using grid search in order to maximise accuracy of each algorithm. The models were used on the same preprocessed dataset with extracted features to see the difference in accuracies to determine which one is the best. All of the models in the end came out with similar results but XGBoost was able to outperform the other models.