Conventional livestock monitoring systems often lack the detail and flexibility needed to capture complex animal behavior patterns. In this work, we propose a multimodal learning approach that fuses data from multiple sensor sources for accurate behavior recognition in dairy cows. We use the MMCows dataset, which provides synchronized recordings from diverse modalities. Our method applies supervised learning to individual sensor streams and employs data fusion techniques to combine complementary information for improved behavior classification. We evaluate our multimodal model to measure its generalizability and robustness, achieving the highest F1 scores across all behavior categories. Our work demonstrates the effectiveness of multimodal data fusion in advancing livestock behavior monitoring and supports the development of automated, real-time management systems.