📰 Fake News Detector: Binary Classification Model

Libraries: scikit-learn, XGBoost, matplotlib, pandas, numpy
Dataset: ISOT Fake News Detection Dataset

In this project we use the 🐍 Python libraries scikit-learn and XGBoost to build a machine learning model that classifies news articles as fake or real. We combine classical machine learning techniques with engineered textual features to improve model generalisability and performance.

🧠 Approach

• Text vectorisation: Bag of Words (BoW)
• Feature engineering: % of special characters & % of capitalised characters
• Baseline model: DecisionTreeClassifier with GridSearchCV
• Ensemble model: XGBClassifier with RandomizedSearchCV
• Robustness: Removed dataset-specific artefacts (eg. reuters) from BoW to improve generalisability
  

✅ Results

• 🤖 XGBoost ensemble achieved ~99.8% accuracy, precision, recall, and F1 score
• Top feature: headline_capitalised (engineered)
• Fun insight: second most important vectorized word for classification — "Trump" 🇺🇸

🔭 Future Work

• Test on more diverse, real-world datasets
• Experiment with advanced text vectorisation (eg. word embeddings, transformer models)
• Compare with alternative classifiers (eg. Support Vector Machines)
  

📖 Jupyter Notebook: GitHub | CoLab | Kaggle