Ensemble-Based Machine Learning Approach For Fake News Detection On Telegram With Enhanced Predictive Accuracy

Abstract

The rapid proliferation of fake news on social media platforms has raised significant concerns about misinformation, particularly on messaging applications like Telegram. This trend poses a severe threat to public trust and social harmony. Detecting fake news in such environments requires the development of efficient machine learning (ML) models that can accurately identify misleading content while minimizing false positives and negatives. This research aims to propose a robust machine learning-based framework for detecting fake news on Telegram by analyzing text content and user interaction patterns. Data collection involved scraping a dataset from publicly available Telegram channels, which include both genuine and fake news articles with relevant metadata such as user reactions and engagement levels. To address the problem of fake news detection, a set of machine learning algorithms, including XGBoost, K-Nearest Neighbors (KNN), Decision Trees, and Naive Bayes, were explored. A novel ensemble-based approach, termed Ensemble Feature Fusion (EFF), is introduced, combining the strengths of multiple classifiers to enhance predictive accuracy and robustness against diverse fake news characteristics. Performance metrics such as Accuracy, Engagement-Weighted Accuracy (EWA), False Positive Cost (FPC) , Contextual Precision (CP), and Temporal Consistency Index (TCI)  were evaluated in this research. Results indicate that the proposed model outperforms conventional ML techniques, demonstrating improved classification accuracy and reduced error rates in detecting fake news. This approach provides a promising solution to the growing problem of misinformation on Telegram.