Algorithmic Forecasting in Architectural Pedagogy: A Longitudinal, Explainable Machine Learning Framework for Architectural Design Studio Performance

The architectural design studio serves as the core of architectural education, characterized by project based learning and subjective evaluation rubrics. Identifying struggling before the semester begins remains a critical challenge for educational institutions. Despite advancements in educational data mining, predicting performance specifically within architecture education is rarely tackled in the literature. This study presents a data driven early warning system tailored to architectural pedagogy,

specifically aimed at predicting struggling students in the architectural design studio. Utilizing a longitudinal dataset of 670 student records collected from a higher architecture education institution in Morocco between 2021 and 2024, the research incorporates foundational demographics, pre-enrollment metrics, prior academic trajectories, and behavioral indicators. To capture the dynamic nature of student progress, we introduce extracted features representing the trend, volatility, and average of prior studio performances. The Random Forest classifier demonstrated superior performance based on Balanced

Accuracy. Global interpretability analysis using SHAP reveals the overwhelming predictive weight of prior architectural design studio performance, theoretical and technical modules, and historic absenteeism, objectively validating the necessity of continuous engagement in the design process.