The rapid dissemination of fake news through digital media requires accurate and computationally efficient automated detection. This study analyses and classifies fake news using data-mining techniques within the Knowledge Discovery in Databases (KDD) framework. The Fake and Real News Dataset obtained from Kaggle contains 44,898 news articles, comprising 23,481 fake articles and 21,417 real articles. During data preparation, duplicate, incomplete, and irrelevant records are examined; the title and article text are combined into a single content field; and the text is transformed into numerical features using Term Frequency-Inverse Document Frequency (TF-IDF). The corpus is divided into 80% training data and 20% test data and classified using Multinomial Naive Bayes. Evaluation using a confusion matrix and standard classification metrics yields an accuracy of 93.72%, precision of 93.68%, recall of 93.11%, and an F1-score of 93.40%. The model correctly classifies 4,427 fake articles and 3,989 real articles, while misclassifying 269 fake articles and 295 real articles. The balanced class-level results indicate that TF-IDF combined with Multinomial Naive Bayes provides an effective and efficient approach to automated fake-news detection.
Copyrights © 2026