From Prediction to Intervention: Explainable Learning Analytics for Early Warning in Data-Intensive Construction Management Courses and the Causal Effect of Advisor-Led Outreach
Year: 2026
Type: article
Abstract: Python programming and big data analytics now anchor the digital turn in construction management curricula-and they have quietly become the courses where students fail. Institutional responses usually arrive late and reach few. This two-phase study, conducted at a regional applied university in China, asks whether that failure pattern can be interrupted. Phase 1 built early-warning models from learning management system (LMS) and programming-practicum logs covering 269 student–course observations in seven courses; a gradient-boosted model reached AUC = .806 by Week 4, and its SHAP explanations were compared with the judgments of experienced teachers. Phase 2 then tested what the warnings should trigger: a cluster-randomized trial across seven class sections (N = 203) evaluated advisor-led outreach for flagged students. Failure fell from 13.0% to 6.3% (OR = 0.45, 95% CI [0.17, 1.23]) and examination scores rose (d = 0.29), with the largest gains among middle prior-achievement students-achievement gaps narrowed rather than widened. Model explanation and teacher judgment agreed on most features but diverged in places the teachers themselves judged diagnostically valuable. Taken together, the results offer randomized pilot evidence that explainable early warning can be turned into effective, low-cost pedagogical action.
Authors: Chuxuan Ren, Shiqi Teng, Mengting Yuan
Institution: Sichuan University of Science and Engineering
Language: English
Subfield: Computer Science Applications
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7237337222
