Authors:
Rajendra Prasad Joshi

Addresses:
Department of Computer Science and Technology, University of West Scotland, Paisley, Scotland, United Kingdom.

Abstract:

The accelerated rate of online information has worsened the difficulty in differentiating the real news from misinformation. Although there have been breakthroughs in artificial intelligence, most current fake news detection models have inflated accuracy under idealistic validation conditions. In practice, they do not provide human-friendly transparency. This dissertation proposes the design, development, and testing of an AI-based fake news detector that incorporates Natural Language Processing (NLP), transformer-based models, and human-centred explainability to improve algorithm performance and user confidence. The study is based on a comparative experimental approach, which compares classical machine learning models, such as Naive Bayes, Logistic Regression, and Support Vector Machine (SVM), with deep learning and transformer models such as LSTM and Distill BERT. This study examines the performance of false news detection models using the Kaggle False News Dataset, the LIAR Dataset, and a Nepal-focused corpus. Researchers measure the performance using accuracy, precision, recall, F1-score, ROC-AUC, and Expected Calibration Error (ECE). Researchers install the top-performing model in a Django-based web application that delivers calibrated confidence scores and short explanations. The results show that transformer-based models outperform traditional approaches, and that explainable, user-centred interfaces improve trust and decision-making quality in misinformation detection.

Keywords: Fake News Detector; Natural Language Processing; Naive Bayes; Logistic Regression; Support Vector Machine; Expected Calibration Error; Machine Learning; Deep Learning; Artificial Intelligence.

Received on: 20/06/2025, Revised on: 27/08/2025, Accepted on: 24/10/2025, Published on: 07/03/2026

DOI: 10.69888/FTSIS.2026.000668

FMDB Transactions on Sustainable Intelligence and Security, 2026 Vol. 1 No. 1, Pages: 82–101

• Views : 56
• Downloads : 11

Download PDF