🚀 Toxic Terminator: AI-Powered Toxicity Detection 🛡️

"Purifying Digital Spaces One Tweet at a Time" 🔍✨

📋 Table of Contents

1. 📌 Project Overview
2. 📊 Dataset Information
3. 🧹 Data Preprocessing
4. ⚙️ Feature Extraction
5. 🤖 Model Training
6. 📈 Model Evaluation
7. 💻 Installation
8. 🚦 Usage
9. 🚀 Future Enhancements
10. 🤝 Contributing
11. 📜 License
12. 🙏 Acknowledgements
13. 🚀 Deployment Instructions

---

📌 Project Overview

Toxic Terminator is an ML-powered shield against online toxicity 🛡️. Our solution helps platforms:

✅ Automatically flag harmful content
✅ Improve community moderation
✅ Protect user mental health
✅ Maintain positive digital environments

---

📊 Dataset Information

🔗 Source

+ Kaggle Twitter Toxicity Dataset
- https://www.kaggle.com/datasets/ashwiniyer176/toxic-tweets-dataset

📦 Dataset Structure

Column	Type	Description	Example
Unnamed: 0	int64	Index column (removed)	0
Toxicity	int64	Binary label (0/1)	1 (Toxic)
tweet	object	Tweet text content	"@user This is offensive..."

📊 Class Distribution

print(df['Toxicity'].value_counts(normalize=True))

0    57.4% 🟢 (Non-Toxic)
1    42.6% 🔴 (Toxic)

---

🧹 Data Preprocessing

🔄 Cleaning Pipeline

1. 🗑️ Remove index column
2. 🧼 Handle missing values
3. ✂️ Text normalization:
   • Remove @mentions
   • Strip URLs
   • Eliminate special characters
   • Convert to lowercase
   • Remove stopwords

⚙️ Preprocessing Example

Input:
@user Check this link: http://example.com!!! #toxic

Output:
check link toxic

---

⚙️ Feature Extraction

TF-IDF Vectorization Settings

TfidfVectorizer(
    max_features=10000,       # 🎯 Top 10k terms
    ngram_range=(1, 2),       # 🔠 Uni+Bigrams
    stop_words=stop_words     # 🚫 Filter common words
)

Feature Matrix

Dimension	Training Shape	Test Shape
TF-IDF Matrix	(45396, 10000)	(11349, 10000)

---

🤖 Model Training

Model Architecture

graph LR
A[Raw Text] --> B(TF-IDF Features)
B --> C{MultinomialNB}
C --> D[Toxicity Prediction]

Loading

🏋️ Training Parameters

• Algorithm: Multinomial Naive Bayes
• Train Size: 45,396 samples (80%)
• Test Size: 11,349 samples (20%)
• Serialized As: toxicity_model.pkt

---

📈 Model Evaluation

📊 Performance Metrics

Metric	Score	Visual
Accuracy	95.2%	🟢🟢🟢🟢🟢🟢🟢🟢🟢🟢
Precision	92.7%	🔵🔵🔵🔵🔵🔵🔵🔵🔵
Recall	91.3%	🟡🟡🟡🟡🟡🟡🟡🟡
F1 Score	92.0%	🟣🟣🟣🟣🟣🟣🟣🟣🟣
ROC AUC	0.9719	📈 (See curve below)

🔍 Confusion Matrix

	Predicted 🟢	Predicted 🔴
Actual 🟢	9,823	526
Actual 🔴	465	535

---

💻 Installation

Quick Start

# 1. Clone repository
git clone https://github.com/yxshee/toxic-terminator.git

# 2. Install dependencies
pip install -r requirements.txt

# 3. Run training
python notebooks/model.ipynb

🐳 Docker Setup

FROM python:3.8-slim
WORKDIR /app
COPY . .
RUN pip install -r requirements.txt
CMD ["python", "app.py"]

---

🚦 Usage

Real-Time Prediction

from toxic_detector import ToxicityClassifier

detector = ToxicityClassifier()
tweet = "@user You're completely worthless!"
result = detector.classify(tweet)

print(f"🔍 Result: {result['label']} (Confidence: {result['probability']:.2%})")

Output:
🔍 Result: Toxic (Confidence: 98.72%)

---

🚀 Future Enhancements

• 🌐 Multilingual Support
• 🧠 BERT/Transformer Integration
• ⚡ Real-Time API
• 📱 Mobile Integration
• 🔄 Active Learning Pipeline

---

🤝 Contributing

First Time Contributing? 🎉 Here's How:

1. 🌟 Star the Repository
2. 🍴 Fork the Project
3. 🌿 Create a Feature Branch
4. 💻 Commit Changes
5. 🔄 Push to Branch
6. 🎯 Open Pull Request

---

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

---

🙏 Acknowledgements

Organization	Contribution
Kaggle	Dataset Provision
Scikit-learn	ML Framework
Python	Core Language

---

🚀 Deployment Instructions

To deploy the project on Streamlit:

1. Install the required dependencies:

   pip install -r requirements.txt

2. Ensure that the model files (models/tf_idf.pkt and models/toxicity_model.pkt) are in the project directory.
3. Launch the app with Streamlit:

   streamlit run app.py

4. Open the URL provided by Streamlit (usually http://localhost:8501) in your browser.

---

Made with ❤️ by YXSHEE | 🛡️ Keep Conversations Clean!