With the widespread use of English globally, grammar error correction, as an essential tool for improving language accuracy, has gained increasing attention. Traditional grammar correction methods often rely on handcrafted rules and limited corpora, but these methods tend to show limitations when dealing with complex and diverse language errors. As a result, automated grammar correction techniques based on deep learning have become a research hotspot in recent years. In particular, the combination of recurrent neural networks (RNNs) and attention mechanisms has provided new solutions to this problem. This paper proposes an innovative English grammar error classification algorithm based on RNNs and attention mechanisms. We transform the grammar correction task into a classification problem and introduce two attention mechanisms in the model to effectively extract contextual information of target words from input text. The RNN processes and learns this information to identify and correct grammar errors. Unlike traditional methods, our model adopts an end-to-end training approach, leveraging large-scale unlabeled text data through unsupervised learning to supplement the lack of annotated data. In the experimental section, we validate the proposed algorithm using publicly available English grammar correction datasets. The results show that our algorithm outperforms existing deep learning methods in terms of accuracy and error correction effectiveness, demonstrating high precision and strong generalization capabilities, especially when handling different types of grammar errors. Compared with other mainstream methods, the performance of our algorithm is significantly improved, showcasing its potential in practical applications.