Sarcasm is frequently employed in the news, and detecting it in headline news is difficult for humans and hence for computers. The media frequently uses sarcasm in news headlines to capture people's attention. However, individuals find it difficult to identify sarcasm in headline news, so they form an incorrect opinion about it and share it to their peers, coworkers, and so on. The study's goal is to create a sarcasm model that detects news headlines using machine learning and to better understand how a computer learns sarcasm patterns. Therefore, the analysis focuses on what users think about the material, the results offer additional insight into how marketers might organize and present communication content to encourage favorable interaction behaviors. Therefore, it is essential to have a smart system that can recognize sarcasm from non-sarcasm automatically. This model involves four main algorithms: linear SVC, logistic regression, stochastic gradient descent, and XGBoost. The experiments carried out on dataset contains 3000 + sentences, experimental results indicate that logistic regression outperformed among four algorithms with accuracy of 91%.