📰 News Category Classifier

A NLP system that classifies news articles into 16 categories using TF-IDF and Logistic Regression, optimized for fast, real-time inference.

Features

• 16 Categories: Business, Politics, Sports, Tech, Wellness, Entertainment, Travel, Food & Drink, Science, Crime, Environment, Media, Education, Style & Beauty, U.S. News, World News
• Real-time Prediction: Instant classification with confidence scores
• Modern UI: Clean Streamlit interface with responsive design
• Fast Processing: <100ms prediction time on local CPU inference

Quick Start

Local Development

# Install dependencies
pip install -r requirements.txt

# Run locally
streamlit run app.py

Deploy to Streamlit Cloud

1. Push to GitHub:

   git add .
   git commit -m "Initial commit"
   git push origin main

2. Deploy on Streamlit Cloud:

   • Go to share.streamlit.io
   • Connect your GitHub account
   • Select repository: Keerthi421/News_Category_Classification
   • Set main file: app.py
   • Click "Deploy"

3. Access your app at the provided URL

📊 Model Details

• Algorithm: Logistic Regression (Multinomial)
• Features: TF-IDF vectorization (10K features, unigrams + bigrams)
• Dataset: 126K news articles from Kaggle News Category Dataset v3
• Training: 80/20 split, 2-3 minute training time
• Model Choice Rationale: Logistic Regression chosen for interpretability, speed, and ease of production deployment

🤔 Why Logistic Regression?

Logistic Regression provides a strong balance of accuracy, interpretability, and inference speed for large-scale text classification, making it well-suited for production NLP systems where latency and reliability matter.

🛠️ Tech Stack

• ML: Scikit-learn, NLTK
• Web: Streamlit
• Data: Pandas, NumPy
• Language: Python 3.13

📁 Project Structure

├── app.py                 # Streamlit web app
├── train_model.py         # Model training script
├── convert_to_csv.py      # Data preprocessing
├── requirements.txt       # Python dependencies
├── news_data.csv          # Sample / processed dataset
├── model.pkl             # Trained model
├── vectorizer.pkl        # TF-IDF vectorizer
└── label_encoder.pkl     # Category encoder

🎯 Usage Examples

Input	Category	Confidence
"Stock markets decline in early trade"	BUSINESS	34.5%
"New vaccine shows promising results"	WELLNESS	45.2%
"President announces economic policy"	POLITICS	67.8%

🔧 Requirements

• Python 3.8+
• Streamlit
• Scikit-learn
• Pandas
• NLTK

📈 Performance

• Accuracy: 78%
• Designed for balanced performance across classes rather than overfitting top categories
• Best Categories: Politics (91%), Entertainment (84%), Wellness (88%)
• Training Time: ~3 minutes
• Model Size: ~50MB (compressed)

---

Built with a production mindset — fast inference, interpretable models, and clean deployment.