Unsupervised Learning and Parallel Programming With Openmp for Educational Analytics: University Dropout Risk Profiles Using Synthetic Data

Student dropout rates in Colombian higher education exceed 40% in undergraduate programs, highlighting the need for tools to characterize students’ academic trajectories and inform student support strategies. In this context, this study presents the parallel implementation of the K-Means++ and DBSCAN algorithms using OpenMP as a practical teaching experience that integrates educational analytics and parallel programming in computer science the implementation uses 5,000 synthetically generated records, taking the University of Pamplona only as a contextual reference and not as a source of real student data, with academic and sociodemographic variables scaled using the Min-Max method the hyperparameters were selected using the elbow method and the silhouette coefficient, yielding (k = 4) for K-Means and (eps = 0.22) and (min_pts = 3) for DBSCAN, with corresponding silhouette coefficients of 0.6256 and 0.6446. In terms of computational performance, execution with eight threads achieved speedups of 4.11× for K-Means and 4.07× for DBSCAN, while both implementations achieved 100% agreement with scikit-learn furthermore, K-Means revealed four descriptive student profiles high risk, moderate risk, economic risk, and low risk which illustrate the potential of clustering to explore academic and sociodemographic patterns overall, this experience demonstrates the educational value of combining clustering algorithms with parallel programming, while the synthetic nature of the dataset imposes an important limitation: the resulting profiles do not constitute evidence of actual student dropout risk and must be validated against institutional data before being used to inform student support decisions.