Xu et al. proposed a binary classifier that combines sequence three-way decision (S3WD) with classical rough set (CRS), and demonstrated its promising application prospects in binary classification. However, the reference tuples based on CRS become fixed once generated from the training set, which limits the classifier's adaptability to different data distributions; moreover, its binary classification architecture cannot directly handle multi-class classification tasks. To address these issues, this paper proposes EM-S3WD, an integrated multi-class classification framework based on neighborhood rough set (NRS) and S3WD. This framework uses NRS to construct adaptive reference tuples and combines the "one vs all" (OVA) decomposition strategy to construct multiple S3WD classifiers based on NRS. Additionally, a conditional dynamic integration strategy is introduced to achieve the extension of the binary classifier to multi-class scenarios. Finally, experimental results on public benchmark datasets show that the proposed method improves the adaptability of reference tuple construction and achieves competitive classification performance in multi-class classification tasks.