<p>Sentiment analysis becomes significantly challenging when there are traces of sarcastic words or phrases present in text. A common finding of the researchers working on sentiment analysis is that the presence of sarcastic comments escalates the complexity of sentiment analysis drastically. Consequently, it is a common practice to detect the sarcastic elements in a text prior to performing the sentiment analysis so as to increase the accuracy of the same. Sarcastic phrases often have an intrinsic linguistic pattern which the research community is working on during the last few years with an objective to identify them in a generically. But most of the cases the researchers used either rule based or machine learning-driven approach for the detection of sarcasm present in text. The scope of application of deep learning for identifying sarcastic comments is principally due to the scarcity of dataset. In order to bridge the gap, this article presents a novel hybrid transformer based approach that leverages the strengths of RoBERTa, Bidirectional Long Short-Term Memory (Bi-LSTM), and Multi-Head Attention in order to enhance the efficiency of sarcasm detection especially from News Headlines. Our approach combines contextual embeddings, sequential modelling and attention mechanisms to effectively capture the nuances of sarcastic expressions in text. Experimental results demonstrate that our model achieves competitive accuracy and efficiency, providing a robust solution for sarcasm detection tasks.</p>