Classify Real or Fake Disaster Tweets

• Tech Stack: SpaCy, PyTorch, NLTK, Scikit-learn, Pandas
• Github URL: Project Link

Developed models to classify whether tweets were discussing real or fake disasters using three machine learning approaches: Long Short-Term Memory (LSTM), Recurrent Neural Networks (RNN), and a plain multi-layered neural network. The project involved extracting sentence-level and word-level embeddings using the spaCy medium-size model, which captured semantic and contextual features of the tweets. Each model was trained to recognize patterns in tweet content, leveraging these embeddings for accurate prediction.

The models achieved varying levels of accuracy, with LSTM leading at 85%, followed by RNN at 80%, and the multi-layered neural network at 71%. A comparative analysis of these results demonstrated the strengths of LSTM and RNN in handling sequential data, while the neural network provided a baseline for performance. The study provided insights into the effectiveness of different architectures for text-based disaster classification tasks.