Sarcasm detection is an application of sentiment analysis that requires sophisticated models and a substantial amount of annotated data, which may not always be readily available especially for low-resource languages like Arabic. Data augmentation is a technique used to address data scarcity by generating synthetic training data without the need for additional collection and annotation efforts. This approach combined with language models enhances model performance for many natural language processing tasks. In this paper, we propose a practical framework to handle data scarcity for automatic sarcasm detection in the Arabic language using EDA techniques for data augmentation and pre-trained models for classification. The experiments are performed on the two datasets ArSarcasm and MSTD. The results demonstrated notable enhancement in model performance, with increase of 13.8% in F1-score for the ArSarcasm-V2 dataset.