In this project I participated in a kaggle competition that analyzes tweets and attempts to correctly identify tweets that signals real disasters from miscellaneous tweets. The project was executed in a colab notebook.
The project implemented three classifier algorithms (XGBoost, Logistics regression and Multinomial Naive Bayes classifiers) and two texts transformers (CountVectorizer and TFIDF Vectorizer) to find techniques that generate a higher predictive accuracy.
The Tasks were broken down into four steps:
- Exploratory Data Analysis
- Data Cleaning
- Modelling and
- Prediction
From the three algorithms used for prediction, Multinomial Naive Bayes classifier combined with count vectorizer performed better than XGBoost and Logistic regression classifiers combined with TFIDF Vectorizer, with an f1 score of 0.6627, 0.3736 and 0.5938 respectively. Thus, predictions on the test dataset was done using the MNB Classifier and achieved a public score of 0.79803 on the leaderboard