Abewas1493/Amharic-News-Classification

Scraped BBC Amharic articles, applied NLP preprocessing, and trained ML models (Logistic Regression, SVM, Random Forest) achieving 89% accuracy in classifying news categories. # BBC Amharic News Classification This project is a machine learning-based text classification system for BBC Amharic news articles. The goal is to automatically categorize news into predefined topics using natural language processing (NLP) techniques. ## Dataset * **Source:** BBC Amharic * **Collection:** Scraped using Python's BeautifulSoup library * **Size:** 3,868 news articles * **Categories:** * ፖለቲካ (Politics) * ጤና (Health) * ስፖርት (Sports) * ቴክኖሎጂ (Technology) * ምጣኔ ሃብት (Wealth & Economy) ## Methodology * **Text Preprocessing:** Tokenization, cleaning, and TF-IDF vectorization * **Train/Test Split:** Standard split of the dataset for training and evaluation * **Algorithms Tested:** * Random Forest Classifier * Support Vector Machine (SVM) * Logistic Regression * Multinomial Naive Bayes ## Results | Classifier | Accuracy | | ------------------- | -------- | | Random Forest | 0.8680 | | SVM | 0.8900 | | Logistic Regression | 0.8875 | | Naive Bayes | 0.8719 | ### Key Observations * **Best performing model:** SVM (Accuracy: 89%) * **High-performing categories:** Sports (ስፖርት) consistently achieved high precision and recall across all models. * **Challenging category:** Wealth & Economy (ምጣኔ ሃብት) had slightly lower recall compared to other categories. ## Tools & Libraries * Python 3.x * `BeautifulSoup` for web scraping * `scikit-learn` for machine learning models and evaluation * `pandas` & `numpy` for data manipulation