Optimizing Machine Learning Models based on Hyperparameter Tuning for Arabic Fake News Detection

The increase of fake news in Arabic social media causes major challenges to information truth and societal discourse. This study presents a comprehensive framework for optimizing machine learning models to detect Arabic fake news through systematic hyperparameter tuning. Eight machine learning algorithms were evaluated including Naïve Bayes (NB), Support Vector Machine (SVM), Logistic Regression (LR), Random Forest (RF), Gradient Boosting, AdaBoost, Extra Trees (ET), and Stochastic Gradient Descent (SGD), applying three optimization strategies: Grid Search, Random Search, and Bayesian Optimization. Our approach combines TF-IDF vectorization for text preprocessing with SMOTE for dataset balancing. Experimental results demonstrate that Bayesian-optimized SVM achieves superior performance with 98.20% accuracy and 99.72% AUC, significantly outperforming baseline models. The findings provide valuable insights for developing robust fake news detection systems for Arabic content.