Collective Intelligence Dynamism and Group Performance Datasets

To address the limitations of traditional virtual agents, such as detachment from physical contexts and lagging collaboration assessment, this study constructs a field-based intelligent agent architecture embedded within physical training spaces. Based on this architecture, we establish a Collective Intelligence (CI) activity representation system, conceptualizing CI activity (Af) into a three-dimensional computable framework comprising interaction viscosity, collaboration uniformity, and coordination flexibility. Leveraging multi-batch empirical data from 186 students, we confirm the existence of a collective intelligence factor (C factor) independent of individual intelligence in engineering collaboration. For data collection, key predictive variables—including co-presence duration, time standard deviation, and team capability increment—were extracted, alongside Intended Learning Outcome (ILO) scores gathered through phased reporting. Using Random Forest to evaluate feature importance, we screened these core variables to construct Multiple Linear Regression (MLR) and Support Vector Regression (SVR) models. This research provides a robust theoretical and methodological foundation for the real-time perception and quantitative prediction of collective intelligence dynamics.