Unsupervised Machine Learning in the Evaluation of Brazilian Postgraduate Programs: Decision Support Via the Crowm Method

Integrating Machine Learning techniques, particularly Principal Component Analysis (PCA), into evaluating educational programs has become increasingly significant. This study uses the novel decision support application Communalities on Ranking and Objective Weights Method (CROWM), which combines PCA, unsupervised Machine Learning, and Multicriteria Decision Analysis (MCDA) to evaluate Brazilian postgraduate programs. Twenty-four courses are analyzed across seven key metrics of Coordination for the Improvement of Higher Education Personnel (CAPES): CAPES Scale, Physical Structure, Faculty Quality, Society Impact, Efficiency, Intellectual Production, and Relevance. The robustness of the methodology is validated through KMO (0.64), Kaiser’s, and Bartlett’s sphericity tests ([Formula: see text]-value of 1.66[Formula: see text]), complemented by sensitivity analysis using MCDA methods. The results show high Spearman correlations with CROWM: 0.89 for CoCoSo, 1.00 for PCA, 0.95 for THOR 2, and 0.81 for WISP, demonstrating strong alignment with well-established MCDA methodologies. CROWM’s limitations include dataset scope, generalizability, reliance on PCA validation, subjective input exclusion, and data structure sensitivity. Beyond the Brazilian context, this methodology provides a versatile framework adaptable to international educational systems, corporate training programs, and public sector evaluations, provided that the analyzed data are quantitative and objectively defined. Its structured, data-driven approach ensures reliability in decision-making across various fields, where statistical validation and objective evaluation criteria are essential.