Machine Reading Comprehension (MRC) in low-resource languages presents challenges due to the scarcity of annotated data leading to poor performance when compared to resource-rich languages such as English. Data augmentation is a standard approach to improve the number of samples of a low-resource dataset to improve the model performance. With an aim to identify the suitable data augmentation approach for Telugu MRC, we employ back translation and paraphrasing techniques to augment the Telugu MRC data samples of the TyDiQA-GoldP dataset. We fine-tuned Transformer-based models mBERT and XLM-R to extract the answer span from the context of the question. Our empirical study reveals that back translation of question and context together has better Exact Match and F1 scores when compared to no augmentation, back translation of question, and paraphrasing of question.