MYGBM/fake_news_detection

Amharic Fake News Detection System. This application leverages NLP to detect whether news in Amharic is fake or real! # 🛡️ Amharic Fake News Detection System A comprehensive machine learning system for detecting fake news in **Amharic (አማርኛ)** text, featuring specialized text preprocessing for Ethiopic script, multiple classification models, a FastAPI backend, and a modern Next.js web interface. Live Demo • Report --- ## � Table of Contents - Overview - Features - Project Structure - Model Performance - Data Pipeline - Installation - Usage - API Reference - Technology Stack - Dataset - Contributing - References --- ## 🎯 Overview The spread of misinformation in Amharic-language social media poses significant challenges to Ethiopian and global Amharic-speaking communities. This project addresses this challenge by providing an end-to-end solution for detecting fake news in Amharic text using classical machine learning techniques. ### Key Highlights - **Language-Specific Preprocessing**: Custom normalization rules for Ethiopic script (Fidel) handling phonologically similar characters - **Multi-Model Ensemble**: Four trained classifiers providing diverse predictions with confidence scores - **Production-Ready API**: RESTful API with CORS support for seamless integration - **Interactive Web Interface**: Modern, responsive UI with real-time analysis and visualization --- ## ✨ Features ### 🔤 Amharic Text Preprocessing - **Character Normalization**: Unifies phonologically equivalent Ethiopic characters (e.g., ሀ/ሐ/ሓ/ኀ → ሀ) - **Noise Removal**: Eliminates URLs, HTML tags, English text, emojis, and special characters - **Geez Number Handling**: Removes traditional Ethiopic numerals (፩፪፫...) - **Text Cleaning**: Handles elongated words, extra whitespace, and leading/trailing spaces ### 🤖 Machine Learning Models | Model | Description | | ----------------------- | ------------------------------------------ | | **Logistic Regression** | Linear classifier, best overall accuracy | | **Decision Tree** | Rule-based class …