Leveraging Big Data Analytics for Combating Fake News: A Supervised Learning Approach to Identifying Misinformation on Social Media

Abstract

The rapid rise of social media has transformed how people consume and share information but has also accelerated the spread of misinformation that undermines public trust, public health, and democratic stability. Manual fact-checking and platform moderation often lag behind the speed of misinformation, highlighting the need for scalable, automated solutions. This study develops a supervised machine learning framework supported by Big Data analytics for fake news detection. Using the ISOT Fake News Dataset of 44,898 labeled articles, we implemented a structured pipeline that included text normalization, tokenization, stopword removal, stemming, and TF-IDF vectorization, followed by training four classifiers: Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost). Evaluation was conducted using a stratified 80/20 train-test split with 10-fold cross-validation, applying Accuracy, Precision, Recall, and F1-score as performance metrics. Results show that ensemble models, particularly XGBoost and Random Forest, consistently outperformed LR and SVM, achieving accuracies near 99% with strong precision and recall across both classes. These findings demonstrate the strength of optimized ensemble methods in detecting misinformation and their scalability for real-world application. Beyond model performance, this work proposes a distributed architecture leveraging Apache Spark for real-time deployment, providing a foundation for practical and scalable misinformation detection systems.