train.csv
andtest.csv
files can be found via https://www.kaggle.com/competitions/nlp-getting-started/data.- Columns in `train.csv' dataset are:
- id
- text
- location
- keyword
- target
- You will be predicting if tweet is a real disaster (1) or not (0).
- Machine learning models such as LightGBM, XGBoost, RandomForest, and CatBoost Classifiers have been used to predict the disaster tweets.
- RandomizedSearchCv is used to tune hyperparameters for models.
- There is a commented out code in jupyter notebook in which you can
combine other features with tf-idf matrix
usinghstack
just in case of use if wanted.
Models | LGBMClassifier | CatBoostClassifier | XGBClassifier | RandomForestClassifier |
---|---|---|---|---|
Accuracy | 0.7634 | 0.7706 | 0.7648 | 0.7873 |
- RandomForestClassifier has demonstrated higher accuracy than rest of the models. Therefore, Test data is evaluated using RandomForestClassifier.