This project focuses on building a machine learning model to predict whether a tweet is about a real disaster or not. In today's digital age, Twitter has become a crucial communication channel during emergencies. The widespread use of smartphones allows individuals to report real-time observations of emergencies, making it a valuable source of information for disaster relief organizations and news agencies.

However, the challenge lies in distinguishing between tweets that genuinely announce a disaster and those that use disaster-related terms metaphorically or in different contexts. For instance:

While the word "DISASTER" is used to express frustration and inconvenience related to heavy traffic, it doesn't imply an actual disaster like a natural calamity. In this context, the tweet is about a common, non-emergency situation. This ambiguity underscores the need for a reliable machine learning model capable of discerning the true nature of tweets during emergencies.

The dataset for this project is obtained from Kaggle and can be accessed directly on Kaggle's NLP Getting Started Competition Data Page.

• Real Disasters (Target = 1): Tweets associated with actual disasters or emergencies.
• Non-Disasters (Target = 0): Tweets that are not related to real disasters.

The Jupyter Notebook contains the entire code for data loading, preprocessing, model training, and evaluation. The key steps include:

1. Data Cleaning and Exploration: Checking for missing values and exploring the dataset.
2. Text Preprocessing: Cleaning the text data by removing URLs, HTML tags, and non-alphabetic characters. Tokenization and removal of stop words are also performed.
3. TF-IDF Vectorization: Transforming the cleaned text data into numerical features using TF-IDF vectorization.
4. Model Building: Creating an ensemble model using three classifiers - Multinomial Naive Bayes, Logistic Regression, and Support Vector Machine.
5. Model Evaluation: Assessing the model performance on a validation set, including accuracy, confusion matrix, and classification report.

The ensemble model achieved a validation accuracy of approximately 81.44%.Detailed metrics, including precision, recall, and F1 score, can be found in the classification report.

The trained model is applied to the test dataset (test.csv), and the predictions are stored in submission.csv.

• there is a potential avenue to improve the model's performance by incorporating additional data.
• exploring advanced NLP techniques and fine-tuning the model for improved accuracy.