Pri-MCCD: A Real-World Multimodal Dataset for Analyzing Classroom Climate in Primary School Lessons

Unlike experienced teachers who intuitively perceive classroom dynamics, artificial intelligence still struggles to develop comparable perceptual abilities due to the complex, evolving nature of real classroom climates. Capturing this shared emotional and cognitive atmosphere is essential for understanding teaching effectiveness and student development, and especially challenging in primary school settings, where students' expressions are often nonverbal and volatile. Existing datasets are typically task-specific, single-modal, and collected in controlled environments, limiting their ability to reflect authentic multimodal interactions. Annotations also tend to be shallow and lack labels that represent the broader classroom climate. To address these limitations, we present Pri-MCCD, a multimodal classroom climate dataset built from real-world primary school lessons. It integrates visual and auditory features like body posture and acoustic cues. The dataset contains fifteen 40-minute classes across subjects, and is designed to support research in classroom diagnostics, affect-aware systems, and learning environment analysis.