🔍 Offensive Language Classification

A multi-label classification project to detect various forms of offensive content in user feedback using Machine Learning, Deep Learning, and Transformer-based models.

---

📚 Dataset

• Train: English feedback texts with 6 binary labels:
  toxic, abusive, vulgar, menace, offense, bigotry
• Validation and Test: Multilingual texts
  Only the toxic label is available for evaluation.

---

⚙️ Approaches & Models

✅ Machine Learning

• Models: Logistic Regression (LR), Random Forest (RF)
• Features: TF-IDF, multilingual sentence embeddings
• Tuning: Grid Search & Random Search

🤖 Deep Learning

• Models: LSTM, GRU
• Tokenized and padded sequences with Embedding layers

🧠 Transformer Models

• Models: BERT, XLM-RoBERTa
• Fine-tuned with dropout + dense layers

---

📈 Results Summary

Model	Accuracy	F1-Score
LR	✅ High	⚠️ Moderate
RF	❌ Low	❌ Low
LSTM	✅ High	✅ Better
GRU	✅ High	✅ Better
BERT	✅ High	🏆 Best
XLM-R	✅ High	🏆 Best

---

🛠️ Steps to Run the Code

1. Clone the Repository:

   git clone https://github.com/Sajid064/Offensive-Language-Classification.git
   cd Offensive-Language-Classification
   

2. Install dependencies:

   pip install -r requirements.txt
   

3. Run the notebooks:
   • Execute cells in task/model1_implementation.ipynb
   • Execute cells in task/model2_implementation.ipynb

---

⚠️ Key Challenges and Observations

• Class Imbalance: High accuracy but relatively lower F1-score
• Train-Test Mismatch: Language difference affected performance
• Label Availability: Only toxic label provided for test set