Imbalanced data presents a significant challenge in educational data mining and processing. Existing research and applications, such as predicting student dropout risk, identifying struggling students, or classifying learning behaviors, often involved highly skewed class distributions. The minority class is frequently more critical, yet it is often overlooked by machine learning models due to bias towards the majority class. This research aims to evaluate the effectiveness of data augmentation techniques in addressing imbalanced data in education, particularly in improving classification performance for minority classes. The objective is to develop approaches that enable educators to identify and support at-risk students more efficiently. We employ data augmentation techniques like oversampling and VAE-GANs on real-world educational datasets. Classification models of machine learning and deep learning are trained and assessed using metrics appropriate for imbalanced data, such as F1-score accuracy, confusion matrix, and standard deviation. Our experiments show that data augmentation, particularly with VAE-GANs and oversampling, significantly improves the classification performance of minority classes without compromising the performance of majority classes. Careful selection of appropriate augmentation techniques based on data characteristics is crucial. Our detection also revealed that VAE-GANs consistently demonstrated superior performance compared to oversampling, achieving a remarkable 82% success rate. This study affirms the benefits of using data augmentation in education to handle imbalanced data, thus enhancing the ability to detect and support at-risk students, promoting a more equitable, effective learning environment, and developing more advanced techniques for personalizing learning based on education data.