Time-series data is ubiquitous across various domains, from healthcare and finance to industrial monitoring and human activity recognition. The accurate classification of such data is crucial for informed decision-making and automated systems. However, a common challenge in developing robust time-series classification models, especially deep learning-based ones, is the scarcity of sufficiently large and diverse labeled datasets. Data augmentation has emerged as a powerful technique to address this limitation by synthetically expanding the training data, thereby enhancing model generalization and reducing overfitting. While data augmentation has been extensively studied in domains like image processing and natural language processing, its application and effectiveness in time-series classification present unique challenges and opportunities. This article provides a comprehensive survey of existing data augmentation techniques specifically tailored for time-series classification. Furthermore, it synthesizes empirical findings from a wide range of studies, discussing the efficacy of different augmentation strategies across various datasets and model architectures. We categorize augmentation methods, analyze their underlying principles, and highlight their impact on classification performance. Finally, we identify current limitations and propose future research directions to foster the development of more effective and universally applicable time-series data augmentation methodologies.