Detecting sarcasm, though challenging for machines, is crucial to digital media monitoring and analytics. It plays a crucial role in several applications like digital assistants, analyzing sentiment in product reviews, customer feedback and conversational AI systems. This research attempts to enhance sarcasm detection performance by employing a blended ensemble of four lightweight machine-learning models. Experimental analysis on a sarcasm dataset of News headlines evaluates the performance using metrics: accuracy, precision, recall and F1-score with different combinations of these four components as base learners and meta learner implementing two feature extraction techniques: Bag of Words and Term Frequency-Inverse Document Frequency Vectorization. The best performance is attained using TF-IDF Vectorization with Random Forest as meta learner and Logistic Regression, Passive Aggressive classifier and Multinomial Bayes model as base learners. The proposed ensemble achieves an accuracy of 88% requiring minimal computational resources and execution time thereby demonstrating its effectiveness.