• L. Malarvizhi

Fake news detection remains challenging because high benchmark accuracy can reflect dataset-specific stylistic or source cues rather than robust credibility assessment. This study presents a leakage-controlled hybrid text-classification framework that combines word-level and character-level TF-IDF representations with a Linear Support Vector Machine (SVM). The system was evaluated on the ISOT Fake News Dataset after cleaning, removal of exact duplicate texts, exclusion of source/subject/date metadata from predictive features, and removal of explicit Reuters tokens intended to reduce a known source shortcut. After preprocessing, 39,092 articles were retained and divided into 26,582 training, 4,691 validation, and 7,819 final held-out test instances. Four models were compared on the validation set: Logistic Regression, Linear SVM, Multinomial Naive Bayes, and the proposed Hybrid Word-Character SVM. The hybrid model achieved 99.872% validation accuracy and 99.860% F1-score and was selected before final testing. On the untouched held-out test set, the selected model achieved 99.81% accuracy, 99.83% precision, 99.75% recall, 99.79% F1-score, and 99.81% macro-F1. The implementation also maintains a model version identifier and supports integration with a desktop credibility-analysis application. These results demonstrate strong in-domain discrimination on ISOT; however, because ISOT's real and fake classes originate from systematically different source collections, the paper does not interpret the result as universal factual-verification accuracy. Cross-dataset validation and evidence-grounded fact verification are identified as essential future work.

fake news detection; TF-IDF; character n-grams; support vector machine; explainable AI; credibility analysis; ISOT; machine learning

"A Leakage-Controlled Hybrid Word-Character TF-IDF Framework for Explainable Fake News Detection", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.13, Issue 10, page no.a890-a893, October-2026, Available :http://www.jetir.org/papers/JETIR2610090.pdf

"A Leakage-Controlled Hybrid Word-Character TF-IDF Framework for Explainable Fake News Detection", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.13, Issue 10, page no. ppa890-a893, October-2026, Available at : http://www.jetir.org/papers/JETIR2610090.pdf