HMFED: A Hybrid Multi-Feature Ensemble Framework for Real-Time Fake News Detection Using NLP and Machine Learning

Abstract: The rapid proliferation of misinformation across digital platforms has emerged as one of the most pressing sociotechnical challenges of the modern era. Existing detection approaches either depend on resource-intensive deep learning models that require substantial computational infrastructure, or rely on simplistic keyword-based heuristics that fail to capture the semantic complexity of fabricated content. This paper presents the Hybrid Multi-Feature Ensemble Detection (HMFED) framework, a lightweight yet high-accuracy fake news detection system that integrates NLP preprocessing, multi-category feature engineering, and a weighted soft-voting ensemble of three heterogeneous classifiers. The feature extraction pipeline extracts over 10,000 features spanning textual attributes (TF-IDF with bigrams, lexical diversity via Type-Token Ratio), sentiment attributes (VADER compound scores, TextBlob subjectivity), and metadata attributes (source credibility index, author reliability score, and temporal posting patterns). Three classifiers — Random Forest, Gradient Boosting, and Support Vector Machine with RBF kernel — are combined through a cross- validation-weighted soft-voting mechanism. Evaluated on the benchmark LIAR dataset comprising 12,836 PolitiFact- annotated statements, HMFED achieves 97.4% accuracy, 97.1% precision, 97.3% recall, and an F1-score of 97.2%, outperforming BERT Base (94.6%) while requiring no GPU infrastructure. This work addresses a critical gap between computationally heavy, high-accuracy deep learning approaches and lightweight, accessible detection systems, offering a practical solution suitable for deployment in resource-constrained MCA-level research settings and beyond.

Keywords: Fake News Detection, Natural Language Processing, Machine Learning, Ensemble Learning, Feature Engineering, Sentiment Analysis, VADER, TF-IDF, LIAR Dataset, Misinformation