Education arrangements, AI integration, and supervision satisfaction in thesis writing: Pedagogical empirical insights using XGBoost and RF models

Higher education faces persistent tension between fostering student satisfaction and enhancing measurable learning performance. While pedagogical research has long debated the role of teaching methods in shaping thesis writing outcomes, empirical studies in urban planning and design (UPD) education remain fragmented. Existing scholarship often treats satisfaction as a secondary byproduct of instruction rather than as a central determinant of performance. This gap obscures the potential of “UPD education” as both a pedagogical goal and a performance-enhancing mechanism. This study explores how AI integration and thesis supervision shape students' satisfaction within higher education. Satisfaction is decomposed into two dimensions: thesis supervision educational satisfaction (EY1) and overall educational satisfaction (EY2). Using survey data from undergraduates from the UPD department, we integrate ordinary least squares (OLS) regression, random forest (RF) models, mediation analysis, and interpretable eXtreme Gradient Boosting (XGBoost) with SHAP values. Results show: (1) EY1 is primarily driven by efficiency gains—each unit increase in AI application proficiency raises process satisfaction by 0.519 units—while EY2 rises by 0.355 units for every unit increase in schedule rationality. (2) EY1 acts as a full mediator, transmitting AI use and personal factors' positive effects onto EY2. (3) Teaching arrangement quality exerts a steep threshold effect: once it surpasses a critical point, overall satisfaction surges sharply. Theoretically, the study reframes UPD education as a measurable convergence of satisfaction and performance, rather than a diffuse cultural mission. Practically, it provides a framework for recalibrating curricula, striking a balance between affective engagement and academic rigor. By pinpointing which pedagogical factors shape both dimensions, the findings provide actionable strategies for universities seeking to optimize pedagogical workflows, elevate teaching quality, and advance the integrative aims of thesis writing.