Rising incidence of fake news on social media has turned verifying information into an imperative issue; hence, fact-checking information is becoming an important task. The traditional machine learning-based models like Logistic Regression, Naïve Bayes, Support Vector Machines, and Random Forest suffer from the high-dimensional textual data, and the model may not yield optimal results in fake news detection classification. This paper suggests a better detection framework incorporating Gradient Boosting, CatBoost, and AdaBoost, along with Multinomial Naïve Bayes for comparative study. This research uses TF-IDF vectorization and advanced text preprocessing, such as stopword removal, tokenization, and feature engineering,are done for better classification accuracy. The research was carried out on public dataset, including the Fake Job Posting dataset of Kaggle, to ensure model flexibility. The findings show remarkable performance enhancement with CatBoost posting the best accuracy of 98.23% and an ROC-AUC score of 0.9739, surpassing traditional models. A statistical significance test (t-test) validates the improvements as significant. Results have shown that ensemble-based approaches perform well in handling imbalanced and high-dimensional text data, and they should be generalizable to real-world fake news detection tasks.