Dynamic decision-support for engineering course evaluation and improvement: a case study of the electric machinery and drives course.

Engineering education accreditation emphasizes that course evaluation should enable continuous improvement. However, existing evaluation models are often constrained by static characteristics, single-dimensional assessment, insufficient predictive capability, and a disconnected closed-loop optimization mechanism. To address these challenges, this study develops a dynamic educational decision-support framework encompassing evaluation, diagnosis, prediction, and optimization. Using five-period data from 2021 to 2025 (n = 882) for the Electric Machinery and Drives (EMD) course, the framework is empirically demonstrated the proposed framework on a single-course longitudinal dataset, establishing 16 dynamic evaluation indicators covering both student and instructor dimensions. The framework integrates temporally adaptive weighting, correlation-aware multidimensional evaluation, trend-aware shortcoming diagnosis, and uncertainty-aware prediction. Specifically, sliding-window entropy weighting captures temporal changes in indicator importance, Mahalanobis-distance TOPSIS accounts for inter-indicator correlations, Markov prediction with Bootstrap confidence intervals estimates future state distributions, and D-SRC is formulated as a trend-aware extension of the Shortcoming Repair Coefficient to support proactive teaching intervention. On the basis of the three traditional dimensions (importance, deficiency, and student differentiation), the short-term evolutionary trend of indicator scores is introduced to realize bottleneck identification with both status diagnosis and prospective early warning. A three-dimensional, nine-measure optimization path is then formulated. The five-year analysis showed that the weights of technology-driven and ethics-related indicators changed substantially, with b10 and b15 increasing by 228% and 150%, respectively. M-TOPSIS provided correlation-aware evaluation and increased the SD-based relative dispersion of closeness coefficients by 10.7%, while Markov prediction with Bootstrap confidence intervals generated a short-term baseline projection of the 2026 excellent rate. After D-SRC-guided interventions for b15, b10, and b8, the corresponding scores increased by 7.7, 7.3, and 6.3 points, and the overall excellent rate increased from 24.1% to 39.6%. The proposed framework provides a decision-support tool for engineering education continuous improvement. By incorporating temporal trend information, D-SRC extends static shortcoming diagnosis toward prospective early warning and supports targeted intervention planning. The three-dimensional, nine-measure strategy further links diagnosis with teaching improvement, supporting the transition from static course assessment to dynamic, feedback-oriented educational decision support. Future studies should validate the framework across additional courses and institutions.