Real-Time Fake News Detection in a Social Networking Platform Using Ensemble NLP: A Deployable Approach for African Contexts

The proliferation of fake news on social media platforms poses a growing threat to information integrity, political stability, and public trust, particularly in the African context where localised fact-checking infrastructure remains limited. This paper presents the design, implementation, and evaluation of a real-time fake news detection system embedded within an AI-moderated social network. The proposed system employs an ensemble machine learning approach, combining Logistic Regression, Random Forest, and Extreme Gradient Boosting (XGBoost) models trained on a corpus of 20,720 labelled news articles. Natural Language Processing (NLP) techniques, including tokenisation, lemmatisation, stop-word removal, and TF-IDF vectorisation, are applied to transform raw textual data into discriminative feature representations. Experimental results demonstrate strong predictive performance, with XGBoost achieving a peak Area Under the Curve (AUC) of 99.69% using character-level feature representation. The system was further evaluated on ten real-world news articles, including content sourced from Zimbabwean news platforms, achieving an average accuracy of 94.09% and providing preliminary evidence of cross-domain generalisation. The system accepts news input in multiple formats (plain text, URL, and PDF) and delivers real-time predictions with interpretable visualisations through a progressive web application. This work contributes a practical and deployable fake news detection framework suited for resource-constrained environments in sub-Saharan Africa, and demonstrates the viability of ensemble NLP methods for combating misinformation at scale. Keywords— Artificial intelligence moderation; Ensemble learning; Fake news detection; Misinformation; Natural Language Processing; Social media; Sub-Saharan Africa