Campus planners still rely on surveys and static indicators that miss high-frequency behavioral change. The study developed a Space–Behavior–Decision framework for Wenzhou-Kean University using 243,062 anonymized half-hourly Wi-Fi records (July 2023–March 2025), road-integration for 30 buildings, and functional labels for 35 access points. Seven regression models were compared. Random forest achieved the highest accuracy (R² = 0.9998, RMSE = 0.0025, MAPE = 0.38 %). The decision tree remained nearly as accurate (R² = 0.9986, MAPE = 0.47 %) while offering transparency and easy GIS integration, making it preferable for routine planning. Feature importance gave weights of 0.48 (density), 0.32 (integration), and 0.20 (function). The high fit mainly shows the composite index is learnable. Quarterly updating and post-occupancy checks are required before operational use.