Toxic Comment Classification 🚨

A machine learning pipeline for detecting toxic comments in text using LSTM-based neural networks. This project demonstrates text preprocessing, tokenization, and sequence modeling for text classification tasks.

🔹 Highlights

• Cleaned text data using Texthero (removed digits, URLs, stop words, extra spaces).
• Applied tokenization and padding to standardize input lengths.
• Built an LSTM model with dense, batch normalization, and dropout layers to prevent overfitting.
• Supports predictions on training, and test datasets.
• Multi-label classification for detecting multiple toxicity types per comment.

✨ Features

• Text Cleaning Pipeline: Cleans input text using Texthero, removing digits, URLs, extra spaces, stop words, and other noise.
• Tokenization & Padding: Converts cleaned text into tokens and applies padding to ensure uniform input size for the model.
• Deep Learning Model: Builds a classification model using:
  • Cell-state LSTM layers for sequence learning
  • Dense layers for feature extraction
  • Batch normalization to stabilize training
  • Dropout layers to prevent overfitting

🛠 Tech Stack

• Programming Language: Python
• Libraries / Frameworks: Texthero, TensorFlow / Keras, NumPy, Pandas
• Model Type: LSTM-based Neural Network for text classification
• Environment: Jupyter Notebook

🚀 Setup & Installation

1. Clone the repository

git clone https://github.com/YoshaM09/Toxic-Comment-Classification.git
cd Toxic-Comment-Classification

2. Install dependencies

pip install -r requirements.txt

3. Open the notebook

jupyter notebook notebook1_toxic_comment_classification.ipynb
jupyter notebook notebook2_toxic_comment_classification.ipynb

4. Run the Notebook

• Run the cells sequentially to reproduce the text preprocessing, tokenization, and model training steps, and to generate predictions on both training, and test datasets.

🎬 Usage

• Input raw text comments into the notebook.
• The model outputs predictions for the following categories: toxic, severe toxic, obscene, threat, insult, identity hate.
• Each comment can belong to one or more categories, enabling multi-label classification.

🤝 Contributing

• Contributions are welcome! Please submit a pull request or open an issue for suggestions.

📄 License

• This project is licensed under the MIT License.