Sarcasm is a complex mode of communication that can deceive and mislead analytical systems. Due to its ambiguous nature and being not straightforward identifying sarcasm in text becomes a challenging task. Recent techniques in machine learning based on feature extraction have shown promising results, but they lack in providing novel features such as pragmatics and similarity measures which clearly detects the underlying meaning of the text, thus helping in reducing ambiguity. In addition, understanding the internal working of the model and identify which features are significant an explanation-based system needs to be developed. Therefore, in this research authors propose the two-fold solution. In first fold, authors have identified novel features such as pragmatic and similarity to detect sarcasm and evaluated its performance. In second fold, model-wise explanations are generated with Explainable Artificial Intelligence (XAI) to provide understanding of the model’s outcome with two widely used techniques namely Local Interpretable Model-agnostic Explanations (LIME) for local explanations and SHapley Additive exPlanations (SHAP) for local and global explanations.