Abstract

The existing fake news detection methods mostly rely on traditional machine learning approaches and text-based classification techniques, but many of these methods face limitations in understanding complex linguistic patterns, achieving high detection accuracy, and providing real-time misinformation verification. This paper proposes an AI-powered Neural Network Fake News Detection Framework (NN-FNDF), a deep learning-based system that leverages natural language processing (NLP) and neural network architectures to automatically identify and classify misleading information. The proposed framework performs comprehensive text preprocessing, including normalization, noise removal, stop-word elimination, and stemming, followed by feature extraction using TF-IDF with unigram and bigram representations to generate an 8,000-dimensional feature space. For classification, the framework introduces a progressive feed-forward neural network architecture with multiple hidden layers (1024, 512, 256, and 128 neurons), batch normalization, staged dropout regularization, and sigmoid-based binary classification. The model is optimized using the AdamW optimizer with adaptive learning rate scheduling and gradient control strategies to improve stability and prevent overfitting. Furthermore, the system incorporates a user-friendly real-time detection interface that provides prediction results, confidence scores, and text analysis information for end users. Experimental evaluation on a dataset containing 6,335 real and fake news articles demonstrates that the proposed framework achieves 92.66% accuracy and a 0.9816 ROC-AUC score, outperforming conventional approaches such as Naive Bayes and K-Nearest Neighbors. The proposed framework provides a scalable and efficient solution for journalists, educators, social media platforms, and fact-checking organizations to combat misinformation. Future work will focus on integrating transformer-based models, explainable artificial intelligence (XAI), multimodal information analysis, and adaptive learning mechanisms to improve robustness against evolving misinformation strategies.

Authors

Farhan Shah, Muhammad Rashid Majeed, Ayan Ali
Nanjing University of Information Science and Technology, China

Keywords

Fake News Detection, Deep Learning, TF-IDF, Neural Networks, Misinformation