Multi-class building change detection in urban areas is critical for urban monitoring, disaster management, and planning. This study focuses on improving multi-class change detection performance in high-resolution aerial imagery by addressing the challenges of data imbalance and leveraging a Visual State Space Model (VSSM)-based architecture. To mitigate class imbalance, we introduced object-based random cropping and dynamic class weighting, which enhanced the representation of underrepresented classes during training. These strategies, combined with the robust spatio-temporal modeling capabilities of VSSMs, enabled accurate classification of changes such as new construction, demolition, renovation, color change, and no change. The proposed approach achieved balanced performance across all categories, with a mean Intersection over Union (mIoU) of 0.575 and Cohen's Kappa Coefficient (Kappa) of 0.691, demonstrating its effectiveness in multi-class building change detection tasks in complex urban environments.